Home Improvement Call Center Script Starts With Who Dialed

TL;DR: A home improvement call center script is two different legal animals depending on who placed the call, and most template libraries never mention it. The federal telemarketing rule exempts calls where the sale is not completed and no payment is taken until after a face to face presentation, which is exactly the appointment setting model, but four requirements survive that exemption: no threats or intimidation, caller ID transmission to the homeowner, the do not call provisions, and outbound calls to a residence only between 8 a.m. and 9 p.m. local time at the called person’s location. Inbound calls a homeowner places after seeing your ad are exempt too, right up until the agent upsells, and the exemption stops covering that upsell. Separately, the telephone rules require a written do not call policy available on demand, trained agents, every do not call request recorded at the moment it is made and honored within no more than ten business days, and the individual caller’s name plus a reachable number that is not a 900 number. Then the appointment pays off in the living room, and a contract signed there for $25 or more carries a right to cancel until midnight of the third business day, where business days exclude Sunday and federal holidays, the buyer has to be told orally, and the paper cannot be sold to a finance company before midnight of the fifth business day. So the script’s job is to book a qualified visit, not to pre close one. Kixie’s PowerDialer auto dials up to 4 numbers in parallel and its automated lead caller prompts an agent to call a new lead as soon as a web form comes in.

Most home improvement call centers write the script backwards. Somebody drafts a warm opener, bolts on five qualification questions, adds a rebuttal for every objection the floor hears, and ships it. Then the appointment rate looks fine and the cancellation rate quietly eats it. Reps book visits that never happen, estimators drive forty minutes to a homeowner who thought they were getting a phone quote, and signed contracts come apart three days later.

The script is not the problem. A home improvement call center script that nobody mapped to the actual transaction is describing something it does not understand.

What a Home Improvement Call Center Script Has to Do Before It Sells

Start with the actual job. A home improvement call center does not close anything on the phone. It qualifies a property, confirms the work is in scope, and books a human being to stand in somebody’s kitchen. That is the whole product of the phone call.

That matters because the federal rules are built around exactly that split. The telemarketing rule exempts telephone calls in which the sale is not completed, and payment or authorization of payment is not required, until after a face to face sales presentation by the seller. Read that again with your own process in mind. If your agent books an estimate and nobody takes a card number on the phone, most of the rule does not reach the call.

Most. Not all. The exemption carves out four things that still apply, and those four are the spine of any script you write.

So the agent is not selling. The agent is qualifying a visit and setting an expectation the estimator can meet. Every line in a home improvement call center script either moves toward that or gets cut.

Who Dialed Decides Which Rules the Script Follows

Here is the distinction that should sit at the top of your agent training, and almost never does.

Your call center dialed the homeowner. That is an outbound telemarketing call. The face to face exemption applies if the sale closes in the home, but four requirements survive it. No threats, intimidation, or profane language. The telephone number, and the telemarketer’s name where the carrier makes it available, has to reach the homeowner’s caller identification service, with a permitted substitution of the seller’s name and a customer service number answered during regular business hours. The do not call provisions apply in full. And outbound calls to a person’s residence are an abusive practice outside 8 a.m. to 9 p.m. local time at the called person’s location, without that person’s prior consent.

That last one is a dispatch rule, not a script rule. Local time at the called person’s location. A center in Phoenix working an Eastern list is done dialing at 6 p.m. its own clock, and no opener fixes that.

The homeowner called you after seeing an ad. Calls a customer initiates in response to an advertisement through any medium other than direct mail are exempt as well, with named exceptions for investment, debt relief, technical support and certain business opportunity offers that have nothing to do with gutters. One carve out does apply to you: the exemption does not cover any instances of upselling included in the call.

Think about what upselling means on a real inbound call. A homeowner calls about a leaking water heater. The agent handles it, then offers to have the estimator price the furnace while he is there. That second offer is an upsell inside an inbound call, and it falls outside the exemption even though the first half of the conversation did not.

The homeowner called you with no prompting at all. Calls a customer initiates that are not the result of any solicitation are exempt, with the same upselling carve out attached.

Three doors into the same phone room, three different sets of obligations, one home improvement call center script. That is why a script written as pure copy fails. The script has to branch on how the call started, which means your agents need to know, and which means the record has to say.

The Do Not Call Work Happens Before the Script Opens

The telephone rules set a floor your script sits on top of. Any entity making telemarketing calls to a residential subscriber needs procedures for maintaining a list of people who ask not to be called, and those procedures carry four minimum standards.

A written policy for maintaining a do not call list, available on demand. Personnel engaged in any aspect of telemarketing informed and trained in the existence and use of that list. Every do not call request recorded and placed on the list at the time the request is made, honored within a reasonable time that may not exceed ten business days. And identification on the call: the name of the individual caller, the name of the entity on whose behalf the call is made, and a telephone number or address where that entity can be reached, which may not be a 900 number.

The separate telemarketing rule adds the national registry. Calling a number on the Commission’s do not call registry is an abusive practice unless the seller can demonstrate an express written agreement from that person naming the party and the number, or an established business relationship with a person who has not since asked to stop.

None of that is script copy. All of it decides whether the script ever gets read out loud. Kixie’s compliance workflows support checks against the National Do Not Call Registry and internal suppression lists, and the calling practices and legal review stay with the company doing the dialing.

The Inbound Script for a Homeowner Who Called You

Open, find out what broke, confirm you can fix it, book the visit. Four moves.

Agent: “Thanks for calling [Company], this is [Agent]. What’s going on at the house?”

Agent: “Got it. How long has it been doing that?”

Agent: “What’s the address, so I can check that we cover it?”

Agent: “And the best number to reach you at if we get disconnected?”

Agent: “Here’s how it works. [Role] comes out, looks at it, and gives you a written price before anything gets scheduled. I have [Option A] or [Option B]. Which one is easier?”

Notice what is missing. No price. No range. No “typically our customers see.” The agent on the phone has not seen the house, and a number said out loud becomes the number the homeowner remembers when the estimator says something different.

If the project is out of scope or out of the service area, say so in the first ninety seconds. “That’s outside what we do, and I don’t want to send somebody out just to tell you that.” A booked appointment you cannot serve costs an estimator half a day and a review.

And if the conversation turns into an upsell, remember the exemption stops there. Treat that part of the call the way you would treat an outbound one.

The Outbound Script for a Lead Who Asked for a Quote

Outbound to a form fill is still outbound. The agent identifies, states the reason accurately, and checks whether now works.

Agent: “Hi, is this [Name]? This is [Agent] with [Company]. You filled out a form about [project type] on [date]. Is now a bad time?”

Agent: “Takes two minutes. What are you trying to fix or change?”

Agent: “Where’s the property?”

Agent: “Any timeline you’re working against? Permit, closing, insurance, anything like that?”

Agent: “Next step is [approved next step], no charge, no obligation. I’ve got [Option A] or [Option B].”

When the homeowner does not remember filling anything out, do not argue the point. Read back the source and the date if the record has them, offer to take the number off the list, and log the outcome. An agent who wins that argument has just created a do not call request that somebody has to record anyway.

Pace matters more than wording on this call. A power dialer that auto dials up to 4 numbers in parallel changes what an agent does between conversations, which is where most of a shift actually goes. Fewer gaps means less call hesitancy, and call hesitancy is what quietly eats an afternoon.

Qualification Questions That Earn the Appointment

Ask what changes the routing. Nothing else.

Translucent white glass channel branching into five paths on a pale pink ground, four tapering to blunt stubs and one thicker branch continuing to a small rounded pad
  • What are you looking to repair or replace?
  • When did it start?
  • What is the property address?
  • Do you own the home?
  • Is anything actively leaking, sparking, or unsafe right now?
  • When would you want the work done?
  • Who else looks at the quote before a decision gets made?
  • What times are usually open for you?

Eight questions. Every one of them routes the lead, sizes the job, or sets the appointment. Drop any question your dispatch board does not actually use, because every extra one buys you a shorter call and a colder homeowner.

The decision maker question earns its place for a specific reason, and it is not a closing trick. If the person signing is not home when the estimator arrives, the visit produces a second visit. Ask it on the phone or pay for it in windshield time.

The In Home Close Comes With a Cancellation Window

This is the part the script library never covers, and it is the part that decides whether a booked job stays booked.

Translucent white glass slab on a pale pink ground with a rectangular window cut through it, a small glass wedge lifted out above the opening and three shrinking glass beads trailing away

A sale, lease, or rental of consumer goods or services where the seller personally solicits the sale, including sales that follow an invitation by the buyer, and the agreement to purchase is made somewhere other than the seller’s place of business, is a door to door sale once the purchase price hits $25 or more at the buyer’s residence. Your phone booked kitchen table signature is squarely inside that definition. The invitation came from the homeowner and it does not matter.

What follows from that is concrete:

  • The buyer gets a completed receipt or copy of the contract at the time it is signed, in the same language used in the oral sales presentation, showing the date and the seller’s name and address.
  • Right next to the signature line, in bold face type of at least 10 points, a statement that the buyer may cancel the transaction at any time prior to midnight of the third business day after the date of the transaction.
  • Two copies of a completed cancellation notice captioned either “NOTICE OF RIGHT TO CANCEL” or “NOTICE OF CANCELLATION,” in ten point bold face, with the seller’s name and business address and the date by which notice may be given already filled in.
  • The buyer has to be told orally about the right to cancel at the time of signing. Misrepresenting that right in any manner is itself a violation.
  • Business day means any calendar day except Sunday or a federal holiday. Sign on a Friday before a Monday holiday and the clock runs into Thursday.
  • The seller cannot negotiate, transfer, sell, or assign the note or other evidence of indebtedness to a finance company or third party before midnight of the fifth business day after signing.
  • A valid cancellation gets all payments refunded, trade ins returned, and any negotiable instrument cancelled within 10 business days.

There is an exclusion worth knowing, and a trap inside it. A transaction is outside the rule when the buyer initiated the contact and specifically asked the seller to come out to repair or perform maintenance on personal property. But if the seller then sells additional goods or services beyond the replacement parts needed for that repair, those additional goods or services are not covered by the exclusion. The service call stays outside. The replacement system sold during the service call does not.

Now read your phone script again. Every line that creates urgency to sign today is working against a rule that gives the homeowner three business days anyway. “This price is only good while the estimator is standing there” does not shorten the cancellation window. It just gives a hesitant buyer a reason to use it.

The script that holds up says the opposite. The visit is free, the price is written down, and the homeowner has time to think about it. Teams that say that on the phone send estimators into warmer houses.

Voicemail and Text Templates for the Calls Nobody Answers

Most dials do not connect, and the message is where the next conversation starts.

Voicemail, outbound to a lead: “Hi [Name], it’s [Agent] at [Company], calling about the [project type] request on [date]. You can reach me at [number] between [hours]. Again, [Agent] at [number].”

Name, company, reason, number, number again. Nothing else. An agent who improvises a voicemail on every no answer is writing a new one forty times a shift, which is time that could have gone into the next dial. Recording the message once and letting the system drop it is what Voicemail Drop does, and the agent moves to the next call while it plays.

Missed inbound callback: “Hi, this is [Agent] at [Company] returning a call from this number. What’s going on at the house?”

Text after a booked appointment: “[Name], [Role] from [Company] is set for [date] at [time] at [address]. Reply C to confirm or R to reschedule. Reply STOP to opt out.”

Keep the appointment text to confirmation and logistics. A text that starts selling turns a service message into a marketing one, and the consent you have for one is not automatically the consent you have for the other.

Objections Your Agents Will Hear on Every Shift

Clarify, ask one question, offer a proportionate next step. Do not argue a homeowner out of a reasonable position.

“I’m just getting prices.” “That’s fine, that’s what the visit is for. Do you want the written number so you have something to compare, or are you not ready for anybody to come out yet?”

“I need to talk to my wife.” “Makes sense. When is a time you’re both around? That way nobody has to repeat it.”

“How much is this going to run?” “I’d be guessing, and a guess over the phone helps nobody. [Role] measures it and writes the price down while he’s there, and there’s no charge for that.”

“I’m getting three bids.” “Good, you should. What’s going to decide it for you, price or timeline?”

“Just send me something in email.” “I can do that. The email won’t have a price in it though, because nobody’s seen the house. Do you want the email, or do you want the number?”

Every one of those grants the premise first. The homeowner comparing bids is behaving sensibly, and a rebuttal that treats that as an obstacle reads as pressure. Pressure on the phone becomes a cancellation on day two.

What the Agent Logs When the Call Ends

A script with no disposition behind it is a script you cannot coach.

At minimum, every call should leave the direction of the call, how the lead arrived, the outcome, the next step with an owner and a date, and any do not call request with the timestamp it was made. That last one is not optional housekeeping. The rule says the request gets recorded at the time it is made.

Call disposition logging is what turns that from a note somebody meant to write into a field a manager can filter on. Calls, texts, outcomes and recordings logging into the CRM automatically is the difference between a floor you can inspect and a floor you can only guess at.

Then inspect the right number. Not dials. Not even appointments booked. Look at appointments that held, and at the gap between booked and held by agent. One agent booking 14 a week with 4 holding is not outperforming the agent booking 9 with 8 holding. He is generating windshield time.

If you record calls, know which rule you are under. Federal law makes it lawful for a person who is a party to the communication, or who has the prior consent of one party, to intercept it, unless the interception is for a criminal or tortious purpose. Several states require every party to agree, so the recording notice belongs in the opener and not in a policy document nobody reads.

Home Improvement Call Center Script Mistakes That Cost Appointments

Five patterns show up in nearly every home improvement call center script that gets audited on the floor.

Quoting a price to get off the phone. The agent gives a range, the estimator gives a number, and the homeowner hears a bait and switch that was really just a guess.

One script for inbound and outbound. They are different legal animals and different conversations. The homeowner who called you has a problem right now. The homeowner you called is doing something else.

Treating the appointment as the finish line. Booked is not held, and held is not sold. Measure through.

Burying the cancellation right. It is required, it is going to be disclosed, and the homeowner is going to find out anyway. Scripts that treat it as an awkward disclosure teach agents to rush past it, which is exactly the behavior the rule calls out.

No branch for the homeowner who should not be sold. Out of area, renter without authority, project your crews do not do. Give the agent clean language to end it, and give them credit for ending it fast.

Home Improvement Call Center Script Questions Teams Ask

Does the telemarketing rule apply if we only book estimates?

Partly. The exemption covers calls where the sale is not completed and no payment is taken until after a face to face presentation, which fits the estimate model. But the conduct provisions, the caller identification requirement, the do not call provisions, and the 8 a.m. to 9 p.m. residential calling window all still apply to outbound calls.

Do inbound calls from our ads have the same rules?

Calls a customer initiates in response to an advertisement are exempt from the telemarketing rule, with named exceptions and one that will reach you: the exemption does not cover upselling that happens during the call.

How long does a homeowner have to cancel a contract signed at home?

Until midnight of the third business day after the date of the transaction, for a door to door sale of $25 or more at the residence. Business day excludes Sunday and federal holidays, so a Friday signature before a Monday holiday runs the window into Thursday.

Can agents give a price range over the phone?

They can, and most teams regret it. The phone price becomes the anchor, the written price becomes the disappointment, and the cancellation window is already open.

What does the agent have to say about who is calling?

The name of the individual caller, the name of the company on whose behalf the call is made, and a telephone number or address where the company can be reached, and that number cannot be a 900 number.

How many questions should a qualification script have?

As many as your dispatch board actually consumes. Eight is a workable ceiling for home improvement teams booking in home estimates, and anything past that is usually somebody’s old spreadsheet.

Run the Script Against One Week of Recordings

Writing the home improvement call center script is the easy half. Pull twenty recordings from last week, five per agent, and mark four things on each one: whether the agent identified the company and themselves in the first fifteen seconds, whether a price was quoted, whether the next step had a named owner and a date, and whether any do not call request got logged while the caller was still on the line.

That is a one hour exercise and it will tell you more than another rewrite. Fix the first broken behavior you can see on tape. Then check whether appointments held moved, because that is the only number in this whole process that pays anybody.

Sources

Sources verified by the Kixie Research Team on November 16, 2026.

How to Make Outbound Calls as a Financial Advisor

TL;DR: Learning how to make outbound calls as a financial advisor is mostly a question of which rulebook your script lands in, and the trigger is different in each one. FINRA Rule 2210(f)(1) treats an unscripted live call as a public appearance, while 2210(f)(4) makes any script used in connection with that appearance a communication in its own right, categorized by audience, content and use. From there the line is a headcount: Rule 2210(a)(2) calls a written communication reaching 25 or fewer retail investors in any 30 calendar-day period correspondence, supervised under Rule 3110(b), and 2210(a)(5) calls anything reaching more than 25 a retail communication, which needs a registered principal’s approval before the earlier of its use or its filing under 2210(b)(1)(A), plus retention under SEA Rule 17a-4(b) with a copy, the dates of first and last use, and the approving principal’s name. One rep’s call block clears 25 in a day, so the script is a retail communication from the first week. Regulation Best Interest uses a different trigger entirely: Regulation Best Interest attaches the moment the call carries a recommendation to a retail customer, and it demands the disclosure of capacity, material fees and costs, and type and scope of services in writing, prior to or at the time of the recommendation, which a phone call cannot deliver. That single word is the reason the first call is a fit check rather than a pitch. Registered investment advisers hit the same hinge from the other side: the Advisers Act marketing rule excludes extemporaneous, live, oral communications from the advertisement definition, and the SEC’s adopting release states that extemporaneous communications do not include prepared remarks or speeches, such as those delivered from scripts, with the record kept under the advisers books and records rule at 204-2(a)(11)(i)(A) as the script, the preparatory materials, or a recording of the call. Projections are the live edge: the SEC approved SR-FINRA-2026-004 on October 15, 2026, adding Rule 2210(d)(1)(F)(iv), which permits a projected performance or targeted return only with written policies and procedures on audience relevance and disclosure of the criteria, assumptions, risks and limitations.

Most advice about advisor prospecting stops at tonality. It tells you to sound natural, open with a question, and avoid pitching too early. Fine. None of that answers the question a compliance officer will ask on Monday, which is what the script is, who approved it, and where the copy lives.

So start there. The words you say on a first call are not governed by one rule. They are governed by whichever rulebook your firm registered under, and each one uses a different trigger to decide when the call stops being a conversation and starts being a regulated communication. Get the trigger right and the script writes itself around it.

This article explains general sales process and the published rules behind it. It is not legal or compliance advice. Your firm’s compliance team approves your lists, scripts, disclosures, recording practices, calling hours and retention.

Why a Financial Advisor Outbound Call Script Changes the Rules

FINRA Rule 2210(f)(1) describes a public appearance as a seminar, forum, radio or television interview, or other speaking activity that is unscripted and does not amount to a retail communication, institutional communication or correspondence. Speaking off the cuff is held to the content standards in 2210(d)(1) and supervised under written procedures. That is the lighter path.

Then comes 2210(f)(4). Any scripts, slides, handouts or other written materials used in connection with a public appearance are themselves communications, and the firm has to comply with every applicable part of the rule based on that communication’s audience, content and use. Read those two paragraphs together and the hinge is obvious. The moment you write the call down, you have created a second thing to govern.

That matters because a script is not optional on a calling floor. Reps need one. Managers need one to coach against. New hires need one on day three. So the practical question is not whether to have a script. It is what the script becomes once it exists.

Whatever it becomes, the content standards in 2210(d)(1) still apply. Communications must be based on principles of fair dealing and good faith, must be fair and balanced, and must give a sound basis for evaluating the facts. No false, exaggerated, unwarranted, promissory or misleading statement or claim. Paragraph (d)(1)(D) adds balanced treatment of risks and potential benefits, consistent with fluctuating prices and the uncertainty of dividends, rates of return and yield. Paragraph (d)(1)(E) tells you to consider the nature of the audience and pitch the explanation to it.

Paragraph (d)(1)(F) is the one that kills most opening lines. Communications may not predict or project performance, imply that past performance will recur, or make any exaggerated or unwarranted claim, opinion or forecast. An opener built on a number the prospect could hit is a forecast. Cut it.

There is a live exception worth tracking. On October 15, 2026 the SEC approved SR-FINRA-2026-004 (Release No. 34-106675), which amends Rule 2210 to let a member project performance or provide a targeted return for a security, a securities portfolio, or an asset allocation or other investment strategy, subject to conditions. The conditions are in 2210(d)(1)(F)(iv): the member adopts and implements written policies and procedures reasonably designed to make sure the communication is relevant to the likely financial situation and investment objectives of the intended audience, and the member supplies enough information for that audience to understand the criteria and assumptions used and the risks and limitations of relying on the figure. FINRA’s rulebook entry lists the amendment without an effective date, so the operative date arrives with FINRA’s implementation notice. Until then, treat projections in a calling script as off limits.

The 25 Retail Investor Line Every Outbound Call Script Crosses

Here is the part that surprises people. The category your script falls into is decided by a headcount, not by content.

Shallow frosted glass tray split by an upright glass divider, a few loose glass discs on the near side and a dense crowd of violet and pink glass discs on the far side, one disc lifted above the divider, on a pale violet ground.

Rule 2210(a)(2) defines correspondence as any written (including electronic) communication distributed or made available to 25 or fewer retail investors within any 30 calendar-day period. Rule 2210(a)(5) defines a retail communication as the same thing distributed or made available to more than 25 retail investors in that window. And 2210(a)(6) defines a retail investor as any person other than an institutional investor, regardless of whether that person has an account with the firm. Prospects count.

Now do the arithmetic on a normal week. A rep running a list through a dialer reaches more than 25 prospects in a single block. The script is a retail communication by Tuesday.

The two categories carry different obligations, and the gap between them is wide:

  • Correspondence is subject to the supervision and review requirements of Rule 3110(b) and its supplementary material on correspondence review, and is retained under that supplementary material and Rule 4511.
  • Retail communications require an appropriately qualified registered principal to approve each one before the earlier of its use or its filing with FINRA’s Advertising Regulation Department, under 2210(b)(1)(A).
  • Retention for retail communications runs under SEA Rule 17a-4(b), and 2210(b)(4)(A) spells out what the record has to contain: a copy of the communication, the dates of first and last use, the name of the registered principal who approved it and the date of approval, and the source of any statistic, chart or projection used in it.

Read that last bullet again. Dates of first and last use. That is a version-control requirement sitting inside a communications rule, and it means an edit to the opener is not a small thing. The old version has a last-use date. The new version needs its own approval.

Newer firms carry one more step. Rule 2210(c)(1)(A) requires a member, for one year from the date its FINRA membership became effective in the CRD system, to file with the Advertising Regulation Department at least 10 business days before first use any retail communication published or used in electronic or other public media. The rule’s own list of media includes telephone and audio recording. A first-year firm’s calling script is squarely in scope, and 2210(c)(5) requires each filing to carry the actual or anticipated date of first use plus the name, title and CRD number of the approving principal.

So the sequence is approval first, then dialing. Not the other way around. A floor that writes a new opener on Wednesday and runs it Wednesday afternoon has skipped a step that leaves a trail in the call logs.

What Regulation Best Interest Does to a Financial Advisor Outbound Call

Regulation Best Interest ignores headcount. Its trigger is a recommendation.

Under Regulation Best Interest, a broker, dealer or associated person making a recommendation of any securities transaction or investment strategy involving securities, including account recommendations, to a retail customer has to act in that customer’s best interest at the time the recommendation is made. The rule defines a retail customer in (b)(1) as a natural person, or their legal representative, who receives such a recommendation and uses it primarily for personal, family or household purposes. Nothing in that definition requires an existing account. A prospect on a first call qualifies.

Then look at how the Disclosure Obligation is worded in (a)(2)(i). Prior to or at the time of the recommendation, the firm or the associated person provides the retail customer, in writing, full and fair disclosure of the capacity in which they are acting, the material fees and costs that apply to the customer’s transactions, holdings and accounts, and the type and scope of services provided, including any material limitations on what may be recommended. Plus all material facts about conflicts of interest tied to the recommendation.

A phone call delivers no writing. That is the whole mechanism. If the rep recommends on the first call, the written disclosure arrives after the recommendation, which is backwards against a rule that says prior to or at the time of it.

The Care Obligation closes the other door. Under (a)(2)(ii)(B), the rep needs a reasonable basis to believe the recommendation is in the best interest of that particular retail customer based on that customer’s investment profile. The rule tells you what the profile includes in (b)(2): age, other investments, financial situation and needs, tax status, investment objectives, investment experience, time horizon, liquidity needs, risk tolerance and anything else the customer discloses. You do not have any of that ninety seconds into a first call.

So the first call is a fit check. Not because it is polite, and not because somebody’s blog said to build rapport. Because the disclosure cannot arrive in time and the profile does not exist yet.

One more provision speaks directly to how a calling floor gets run. The Conflict of Interest Obligation at (a)(2)(iii)(D) requires written policies and procedures reasonably designed to identify and eliminate any sales contests, sales quotas, bonuses and non-cash compensation based on the sales of specific securities or specific types of securities within a limited period of time. A spiff on one product for one week is the pattern that paragraph names. Pipeline targets and activity goals are a different animal; a product-specific push on a deadline is not.

Outbound Call Rules That Apply to Registered Investment Advisers

Advisers do not answer to FINRA Rule 2210. They answer to the Advisers Act marketing rule, and it arrives at the same hinge from the opposite direction.

The first prong of the advertisement definition in (e)(1)(i) covers any direct or indirect communication an adviser makes to more than one person, or to one or more persons if the communication includes hypothetical performance, that offers advisory services with regard to securities. Then it carves out three things, and the first is the interesting one: extemporaneous, live, oral communications.

The SEC explained the carve-out when it adopted the rule. A phone call or live video communication between an adviser and an investor could qualify for the exclusion. In the same discussion the Commission drew the boundary in one sentence: extemporaneous communications do not include prepared remarks or speeches, such as those delivered from scripts.

Same hinge as FINRA. Different rulebook, different words, identical pivot point.

The adopting release also says what happens next. A live oral communication that is not extemporaneous, but otherwise meets the advertisement definition, is an advertisement, and a record of it has to be kept under the advisers books and records rule at 204-2(a)(11)(i)(A). The Commission listed what that record can be: a copy of the prepared remarks, other written preparatory materials, or a recording of the oral communication.

Which means the script and the call recording are not two separate compliance chores. Either one can be the record.

Once a script makes the call an advertisement, the marketing rule’s seven general prohibitions govern what it may say. No untrue statement of material fact and no omission that makes a statement misleading. No material statement of fact the adviser lacks a reasonable basis for believing it can substantiate on demand by the Commission. Nothing that would likely cause an untrue or misleading implication. No discussion of potential benefits without fair and balanced treatment of the material risks and material limitations. No reference to specific investment advice unless it is presented fair and balanced. No performance results included, excluded or time-framed in a way that is not fair and balanced. And nothing otherwise materially misleading.

That second prohibition is the one to design around. Substantiate upon demand is an evidence standard, not a tone standard. If a line in the opener states a fact, somebody has to be able to produce the basis for it.

What to Check Before an Outbound Call as a Financial Advisor

Everything above decides what you may say. A separate set of rules decides whether you may dial at all, and it does not care that you are registered.

Long frosted glass tube with a rippled glass filter disc seated in a collar two thirds along, a line of small violet glass spheres passing through toward the right while one sphere is diverted out a side opening and rests apart, on a pale violet ground.

FINRA Rule 3230 is the telemarketing rule for member firms. It governs calling hours and the firm-specific do-not-call list, and the published analysis of how it interacts with federal telemarketing rules is worked through in detail in the financial advisor voicemail script breakdown, so this article will not repeat it. Federal telemarketing rules layer the national registry and the calling window on top; the TCPA rules for outbound calling cover that ground.

The operational failure is almost never the rule. It is the timing of the check.

A list scrubbed on Monday is stale by Thursday. A prospect who asked a rep to stop on Tuesday is still sitting in the Wednesday call queue if the request was logged in a note instead of a suppression field. That is not a legal subtlety. That is a workflow with the check in the wrong place, and the way to find it is to walk the path from list upload to dial and ask where the suppression lookup actually happens. Walking lists through a scrub before upload is covered in the guide to scrubbing calling lists.

Kixie puts that lookup in the dialing path rather than in a spreadsheet. National DNC compliance screens numbers against do-not-call data as part of Kixie’s business calling and texting, and PowerDialer advances the approved list so the rep is not choosing the next number by hand. The point is not that the feature exists. The point is where the decision sits: if the suppression check lives inside the dialing step, a rep cannot route around it by working from a personal export, and the manager can prove which list version a given call came from.

Run the pre-call list in this order and most of the trouble disappears:

  • Confirm the script version the rep is working from, and that a principal approved that version.
  • Confirm the suppression check runs at dial time, not at list-build time.
  • Confirm the local time at the called party’s number, not the rep’s.
  • Confirm whether the call will be recorded and whether the consent rule in the called party’s state has been satisfied.
  • Confirm the rep can state where the contact data came from, from the record rather than from memory.

How to Open an Outbound Call as a Financial Advisor

Knowing how to make outbound calls as a financial advisor without rewriting the opener every quarter means building the opener out of things that do not need approval again when the market moves. Identity does not change. A question does not change. A forecast changes constantly, which is exactly why it keeps failing review.

A workable shape:

  1. Identify. Your name, your firm, and the fact that this is a sales call. Say it in the first breath.
  2. Give the reason. One sentence on why this person and not a random number. Tie it to something you can point at in the record.
  3. Ask permission. A real question with a real no available.
  4. Ask one or two discovery questions. Open, factual, nothing that assumes you already know their situation.
  5. Propose one next step. An introductory meeting. Not a product.
  6. Document it. Outcome, what was asked, what was promised, and any contact restriction.

Written out, it sounds like this:

“Hi, this is [name] with [firm]. This is a sales call, and I will keep it to about a minute. We work with [specific group] in [area] on [the planning question you actually handle]. I do not know whether that is relevant to you, which is why I am calling. Do you have a minute for two questions, or should I try another time?”

Read that against the rules. It states capacity. It makes no recommendation, so Regulation Best Interest has nothing to attach to. It projects nothing, so 2210(d)(1)(F) is satisfied without an exception. It asserts no fact that needs substantiating on demand. And it gives the prospect a clean exit, which is the line most openers cut because it lowers the connect-to-conversation rate on paper.

It also has a version number, because every word of it is a script and the script is the artifact under review.

Outbound Call Objections a Financial Advisor Should Not Argue

Objection handling on an advisor call has a hazard that other industries do not share. Some of what sounds like an objection is a regulated event, and treating it as a sales problem converts a lost call into a rule problem.

“I already have an advisor.” Not an objection. A fact about their situation. Ask one question about what they are working on, accept the answer, and move. Arguing that their current advisor is doing it wrong is specific investment advice delivered with no profile behind it.

“I’m not interested.” Clarify once, gently, and listen to the shape of the answer. If what comes back is a request not to be called again, the conversation is over and the only remaining task is logging it in the suppression field. Do not run a rebuttal track against a do-not-call request. That is the single most expensive mistake on this list.

“Send me information.” Fine, but understand what you just agreed to. Whatever you send is its own communication and gets categorized the same way the script did, by audience, content and use. A one-off email to one prospect and a template mailed to two hundred are not the same object under Rule 2210.

“This isn’t a good time.” Ask for a specific time and honor the calling window at their location, not yours. Then actually call then. A missed callback is the cheapest trust you will ever throw away.

“How did you get my number?” Answer from the record. If the rep cannot say where the number came from, the problem is upstream in list governance, and the honest answer is to say you will check and to flag the record before anyone calls it again. Never improvise a source.

Financial Advisor Outbound Call Follow Up and Recordkeeping

Follow-up on an advisor desk is where the recordkeeping rules stop being abstract.

For member firms, 2210(b)(4)(A) wants a copy of the communication, the dates of first and last use, the approving principal’s name and approval date, and the source of any statistic or projection in it, retained per SEA Rule 17a-4(b). For advisers, the advisers books and records rule at 204-2(a)(11)(i)(A) wants a copy of each advertisement, and the Commission said that record can be the prepared remarks, the written preparatory materials, or a recording of the call.

That last option changes how to think about call recording. On most sales teams recording is coaching material. On an advisor desk it can double as the record of the communication itself. Check the consent rule in the called party’s state before you switch it on, because the states split on whether one party or all parties have to agree.

Voicemail follows the same logic. A voicemail left on a prospect’s phone is a communication, and a dropped voicemail is the same communication every time, which is an advantage: one message, one approval, one version. Voicemail Drop makes that consistency the default instead of hoping each rep repeats the approved wording. Text follow-up has its own retention rules, worked through in the financial advisor texting compliance breakdown.

The manager’s version of all this is four questions. Which script version ran on this call? Who approved it and when? Where is the copy? And if the prospect asked us to stop, which field holds that, and does the dialer read it?

Financial Advisor Outbound Call Metrics Worth Tracking

Dial count tells you the floor was busy. It does not tell you what to coach.

Break the result into the states a call actually passes through, and the coaching target appears on its own:

  • Attempts, live connections, and the gap between them by time of day.
  • Calls where the rep got permission to continue, as a share of live connections.
  • Qualified conversations, meaning a stated planning question the rep could name afterwards.
  • Introductory meetings scheduled, and separately, meetings actually held.
  • Outcome by list source and by approved script version.
  • Opt-outs and do-not-call requests, counted and routed to a suppression field.
  • Wrong numbers and data-quality failures, because those are a list problem wearing a rep problem’s clothes.

Script version is the one nobody tracks and the one the rules already make you keep. Put it on the dashboard next to connect rate and you can answer a question most desks cannot: did the new opener actually change anything, or did the list change underneath it?

Skip universal dial targets borrowed from another industry. Advisor list sizes, call windows and documentation load do not match a transactional sales floor, and a number that fits one does not fit the other.

Financial Advisor Outbound Call FAQs

Can a financial advisor make outbound calls to prospects?

Calling is a permitted prospecting channel for registered firms, subject to telemarketing rules on hours and do-not-call lists, your firm’s own procedures, and the communications rules covering whatever you say. The question is never whether calling is allowed. It is whether this number, at this hour, with this script, is allowed.

Does an outbound call script need principal approval?

Under FINRA Rule 2210(b)(1)(A), a retail communication needs an appropriately qualified registered principal’s approval before the earlier of its use or its filing with the Advertising Regulation Department. A script reaching more than 25 retail investors in a 30 calendar-day period is a retail communication under 2210(a)(5), which covers any script in regular use on a calling desk.

Does Regulation Best Interest apply to an outbound call?

It applies when the call carries a recommendation of a securities transaction or investment strategy to a retail customer, as Regulation Best Interest defines that term. A first call that asks questions and proposes a meeting carries no recommendation. The moment the rep suggests a specific product, account type or strategy, the Disclosure and Care Obligations attach, including the requirement that the disclosure be in writing prior to or at the time of the recommendation.

Are registered investment advisers covered by the same outbound call rules?

No. Advisers fall under the Advisers Act marketing rule, which excludes extemporaneous, live, oral communications from the advertisement definition. The SEC’s adopting release states that extemporaneous communications do not include prepared remarks or speeches, such as those delivered from scripts, so a scripted call that otherwise meets the definition is an advertisement and the firm keeps a record of it under the advisers books and records rule at 204-2(a)(11)(i)(A).

How long do you keep a financial advisor outbound call script?

Member firms retain retail and institutional communications for the period required by SEA Rule 17a-4(b), in a format and media that comply with that rule, and the record has to include the copy plus the dates of first and last use and the approving principal’s name and date, per FINRA Rule 2210(b)(4)(A). Correspondence is retained under the Rule 3110 supplementary material on correspondence retention and under Rule 4511.

Start with the script, not the tonality. Decide which rulebook it lands in, get the version approved before anyone dials it, put the suppression check inside the dialing step, and keep the copy where a reviewer can find it. Then go run the call block.

Sources

Sources verified and content reviewed by the Kixie Research Team on November 14, 2026.

Can ChatGPT Agent Make Phone Calls? Not By Itself

TL;DR: Can ChatGPT agent make phone calls? No. ChatGPT agent has no phone line of its own, and nothing in the shipped product gives it one. OpenAI describes an agent as a model, plus tools, plus guardrails and human oversight, and a telephone is simply not one of the tools in that box. ChatGPT Voice is a microphone-and-speaker conversation inside the desktop app. On OpenAI’s always-on dot agent you can press a phone button to call it, while calls the agent starts on its own are still documented as planned for after launch. The thing that actually dials is GPT-Live over a SIP trunk you bring. You buy the number, you supply the provider endpoint, the digest username and password, and the caller number, then you POST to /v1/live/sessions with the destination written in full international format. OpenAI does not sell you a carrier. Ringing caps at three minutes, a connected call caps at two hours, and neither limit is configurable in the request. A 200 response means the session started, not that anybody answered. Then the law shows up. 47 CFR 64.1200(a)(1) bars initiating a call with an artificial or prerecorded voice to a cellular number absent an emergency purpose or the prior express consent of the called party, telemarketing needs prior express written consent under (a)(2), and the FCC has already determined that voice cloning and similar technologies qualify as an artificial voice. Inbound answering sits outside that prohibition, which is the single biggest planning lever in this article. Kixie has no native ChatGPT integration, so teams wire the two together with webhooks, the developer API, Zapier, Make or n8n. That split is the honest architecture: the agent decides, a rep talks.

So you asked ChatGPT agent to call a prospect and nothing happened. That is the expected result. The agent is not broken, you did not miss a toggle, and there is no permission buried in settings that turns this on.

Here is the short version. Can ChatGPT agent make phone calls? Not on its own. It has no number, no carrier, and no audio path to the telephone network, so the request fails at the first hop rather than somewhere clever. Every demo you have seen where an AI dialed a real phone had a telephony provider underneath it, and somebody was paying that provider by the minute.

Why does that matter beyond pedantry? Because it decides what you can build, what it costs to run, who is liable when a call goes wrong, and whether the thing is legal before you place the first dial.

Why ChatGPT agent cannot make phone calls by itself

Start with the definition OpenAI uses, because it explains the gap better than any feature list.

A frosted glass tray on a pale violet ground with four moulded recesses, three holding a glass lens disc, a cube and a standing ring, and the fourth recess left empty.

An agent is an AI system that can work through a task on your behalf, and OpenAI says agents bring together three elements: a model to interpret the task and plan next steps, tools to gather information or take action, and guardrails and human oversight to set boundaries. The model is excellent. The guardrails are real. The question is the middle element.

What tools does it actually have? Browsing, files, connected apps, code. Not a dial tone.

That is the entire answer, and everything below is consequence. An agent can only act through the tools it was given, so asking it to place a call is like asking it to print a document on a printer that was never connected. The intent is fine. The capability does not exist.

ChatGPT Voice is where most of the confusion starts. Why? Because it genuinely sounds like a phone call. It runs in the ChatGPT desktop app, it is powered by GPT-Live, and it lets you talk through ideas and coordinate tasks in Chat, Work, and Codex. You allow microphone access, you pick a voice, you talk. It supports natural turn-taking, so you can interrupt mid-response and steer the task without typing.

That is a conversation with your computer, which is useful on its own terms and completely unrelated to telephony, because nothing in it touches a phone network and the person you actually wanted to reach is not on the other end.

OpenAI’s dot agent gets closer to the thing people imagine, and the gap there is even more instructive. You can create, message, or call your dot in the ChatGPT mobile app, desktop app, or a desktop browser. To call it, you open its conversation and select the phone button. You talk through a request, clarify a decision, or ask for a progress update, and you can keep typing while you talk.

Read the direction carefully, because the entire question turns on it. You are calling it.

The documentation is explicit about the other direction: calls initiated by your dot are planned for after launch. That is OpenAI describing its own most autonomous consumer product, and outbound calling sits on the roadmap rather than in the build. Ending a call ends the voice conversation, though work you already assigned keeps running in the background.

So when a rep asks whether the agent can call a lead back while they finish lunch, the answer is no. Not yet, and not without a completely different stack underneath it.

The phone call is placed by a carrier you bring

Now the part that does work, because there is a real path from an OpenAI model to a ringing phone. So what is OpenAI actually selling you here? The model and the session, and nothing else. It is an API path rather than a chat path, and the difference in effort between those two is the whole story.

Two frosted glass platforms of different shapes on a pale violet ground joined by a single curved glass tube with four small spheres travelling inside it.

GPT-Live connects incoming and outgoing phone calls through a SIP trunk. Your provider exchanges call audio with OpenAI, while your backend handles session configuration, call controls, and business logic. SIP signaling uses TLS and GPT-Live requires SRTP for the call audio.

The load-bearing word in that paragraph is your. OpenAI does not hand you a phone number.

To place an outbound call you use a SIP trunking provider, such as Twilio, and you purchase a number from that provider first. Then you supply the trunk configuration on every single request: a provider endpoint that must use sips: for TLS signaling, digest authentication carrying your SIP username and password, a caller number and a destination number, both written in full international format with the country code. SIP URI destinations are not supported, so a clever routing shortcut is off the table. Your provider also has to support TLS signaling and SDES-SRTP media, and you enable SRTP in their settings before the first call goes out.

Then you send a POST to /v1/live/sessions with your session configuration and a transport type of sip.

What did you just agree to own? Count the list. A carrier contract. A number you pay for monthly. Credentials that have to stay under server-side control, which is exactly why the documentation tells you to keep your OpenAI API key and SIP credentials on your server rather than anywhere near a client. A backend that stays awake and listening. None of that is a prompt, and none of it is something a sales manager can configure on a Friday afternoon.

One detail catches experienced teams out. Outbound SIP calling is available through the Live API, not the Realtime API call-creation endpoint, so if you built against the older surface the method you are reaching for is not the one that dials.

What about calls coming the other way? Inbound runs as the mirror image, and it is meaningfully simpler. You point your provider’s SIP trunk at your project URI, OpenAI fires a live.transport.incoming webhook when a call arrives, and your backend decides whether to accept or reject it. The documentation also tells you to treat the SIP headers as untrusted caller metadata rather than authorization, which is sound practice in any inbound system and considerably more urgent when a language model is about to act on whatever it was told.

The call limits OpenAI bakes into every phone session

This is where a convincing demo and a production deployment separate, and the published numbers are specific enough to plan against.

Ringing is limited to three minutes. A connected call is limited to two hours. Neither limit is configurable in the creation request.

Is three minutes enough? For a single dial, easily. As a retry strategy it is useless, so your queue logic has to live in your code rather than in the session. Two hours is far longer than any qualification call will ever run. Design against both anyway, because a session nobody closed is still a session somebody pays for.

The response behavior is the part that burns people. When you create the session you get a 200 back once it initializes, and the documentation is blunt about what that does not mean: the call has not been answered. The response carries no SDP and no trunk credentials.

So what actually tells you a human picked up? Three events do, and you have to be listening for them. transport.ringing means the provider reports ringing or early media. transport.answered means the call was answered and media is established. transport.failed means setup failed after initialization, and you inspect the error code and message to find out why.

You read those over a sideband WebSocket that you attach to the session after creating it. Attach immediately. The sideband replays only the preceding three seconds of events before it starts streaming new ones, so a late attachment quietly misses earlier call progress and you end up debugging a call you have no record of starting.

So what should you actually trust? Not the status code. That is the full operational shape, and it is worth saying in one line. A dial attempt is not a result, a 200 is not a connection, and the only proof a conversation happened is an event you were already listening for.

Anyone selling you an agent that “just calls people” has skipped every requirement above and handed you the bill for them later.

What the law counts as an AI phone call

Suppose you build the whole thing and it works. Now ask the question that stops most of these projects before launch.

47 CFR 64.1200(a)(1) says no person or entity may initiate any telephone call using an automatic telephone dialing system or an artificial or prerecorded voice to the lines it lists, other than a call made for emergency purposes or made with the prior express consent of the called party. One of those lines, at (a)(1)(iii), is any telephone number assigned to a cellular telephone service, a paging service, a specialized mobile radio service, or any service for which the called party is charged for the call.

Does that reach your list? Almost certainly. It covers most of the numbers sitting in your CRM right now.

The bar rises when you are selling something. Under (a)(2), a call that includes or introduces an advertisement or constitutes telemarketing, placed with an artificial or prerecorded voice to those same lines, requires the prior express written consent of the called party rather than ordinary consent.

Can you argue your way around it? The obvious dodge is to claim a synthesized conversational voice is not an “artificial” one in the statutory sense. That argument is already closed. In its 2024 rulemaking on AI and robocalls, the FCC noted that the TCPA expressly covers artificial or prerecorded voice, and that the Commission has already determined that voice cloning and similar technologies qualify under that statutory phrase.

There is one asymmetry here that is worth designing your entire roadmap around. The same FCC document states that the TCPA’s prohibition on using an artificial or prerecorded voice message extends only to outbound calls that are made or initiated by the caller, and that the requirements do not extend to technologies used to answer inbound calls.

Read that twice. It is the most useful sentence in this article.

Why does the direction matter so much? Because an AI voice answering your main line is a fundamentally different legal object from the same voice dialing out of it. Same model, same vendor, same code, opposite exposure. So if you want AI voice in production this quarter and you would rather not stand up a written-consent program first, the inbound side is where the room is, and that is a planning decision rather than a technical one.

The Commission has also proposed requiring callers who use AI-generated voice to disclose at the beginning of each call that the call uses the technology. Proposed is doing real work in that sentence, so treat it as a direction of travel rather than a rule you can cite today, and have counsel review anything you intend to dial.

Where an agent belongs in a sales call workflow

Strip the hype out and a clean division of labor appears, which is usually what happens once you know exactly what a tool can and cannot reach.

So where does the agent actually earn money? Not on the call. It is genuinely good at the work around it. It reads the account, assembles context, drafts the opener, pulls the last three touches, and writes the follow-up while the rep is still saying goodbye. It never gets tired at dial ninety and it never skips the notes because the next call is already ringing.

It is bad at the call itself. Not technically bad, though latency and interruption handling are real problems. Bad commercially.

Why? Because a prospect who changes tone at minute two needs somebody who can hear it and throw the plan away, and that is the precise moment a scripted system keeps going. Sticking to the script after the buyer has moved is a named failure mode in human reps too. It is simply guaranteed in a system that has no other option.

The practitioner read on tooling is old and still correct. A tool earns its place when it changes what the rep no longer does by hand, when the outcome lands in the system of record automatically, and when a manager can inspect what happened afterward. An agent that drafts follow-up and logs context clears all three of those. An agent that holds the conversation clears none of them and adds a compliance surface you now own.

So model it as a queue rather than a replacement. The lead arrives, the agent enriches and prioritizes it, a rep dials, the outcome is logged against the record, and the agent handles the admin on the back end so the rep is already on the next call. Dead time disappears. Judgment stays with a person.

Here is the diagnostic question to put to any AI calling pitch, and it works on vendors and internal builds alike. When the buyer says something the script did not anticipate, what happens in the next four seconds? If the answer is a warm transfer to a human, you have bought a routing layer with extra steps and a monthly carrier bill. If the answer is that the agent keeps talking, you have bought a liability with a pleasant voice.

Connecting ChatGPT to Kixie calls without a native integration

Worth stating plainly, because it sets expectations correctly and saves a procurement conversation: Kixie does not have a native ChatGPT integration. Teams connect the two with custom builds on the Kixie API and webhooks, or through Zapier, Make or n8n.

That is not a gap to apologize for. It is exactly the architecture the previous section argued for.

Which direction does the data move? Both, and each direction does different work. Kixie webhooks post completed calls, outcomes, recordings, and texts to your automation layer, which is where a ChatGPT workflow picks them up, so the model reasons over what actually happened on the line instead of what a rep remembered to type an hour later. Going the other way, an event in that workflow can trigger a Kixie call or text through the same automation layer rather than a native endpoint, because the Kixie developer API has no direct auto-call endpoint and those recipes are brokered through Zapier. New records can also be pushed into a PowerList or Cadence so reps can start dialing them immediately.

Notice who dials in that sentence. Reps do.

The developer API and webhooks are included on all plans, which matters because this pattern depends on them rather than on a premium tier. Start Call, Answered Call, End Call, Disposition, and Call Outcome are the events your automation listens to, and they are enough to drive a full post-call workflow. Keep the API key server-side, the same discipline OpenAI asks for with SIP credentials.

On the dialing side, the Multi-Line PowerDialer addresses the volume problem that made people want an AI caller in the first place, while keeping a human voice on every connection. Conversation Intelligence then gives the model something real to summarize. One caution that is easy to get backwards: Voicemail Drop messages are recorded by a rep, and a synthetic voice does not belong in that slot.

If you want the broader version of this question beyond the agent product specifically, we covered whether ChatGPT can make phone calls and why your carrier still does the dialing.

What to check before you let an agent make phone calls

Run this list before anyone writes code. It is short on purpose, and most projects die on the first item.

First, name the carrier. If nobody on the project can say which provider holds the trunk and who pays that invoice every month, there is no calling system yet, there is a diagram.

Second, write down the one task. What single job is this for? Appointment confirmation, a reminder, a callback collection, an inbound routing decision. One of them. Open-ended selling is not a first deployment and probably is not a second one.

Third, settle the direction before anything else gets built, because inbound answering and outbound dialing carry genuinely different obligations under 64.1200. Pick one, get it reviewed, then build.

Fourth, decide what the agent is allowed to read. How much of the CRM does this job actually need? A model with your entire CRM in scope will eventually say something from it out loud to the wrong person.

Fifth, define the stop condition. Which sentence from the buyer ends the automation and hands the call to a person, and how many seconds does that handoff take?

Sixth, test the ugly cases rather than the demo path. Voicemail, dead silence, an interruption at four seconds, a disconnect mid-sentence, a wrong number, an accent the model handles poorly, and a connected system that times out while the buyer waits.

Seventh, check what gets logged. Where does the outcome land? If the call does not land in your system of record with an outcome attached, your manager cannot coach it and your compliance team cannot evidence it, which means it may as well not have happened.

Who owns the result when it goes wrong? You do. One more thing sits underneath all seven. OpenAI’s own guidance says an agent does not share a colleague’s responsibility for the outcome, and that you remain responsible for deciding whether the work is ready to use. On a phone call, that decision gets made in real time by whoever is on the line.

Track the carrier, the task, the direction, the data scope, the stop condition, and the log. Now there is something to review.

ChatGPT agent phone call FAQs

Can ChatGPT agent make phone calls to a real phone number?

Not by itself. The consumer product has no telephony tool at all. Placing a real call requires GPT-Live connected to a SIP trunk from a provider you contract with separately, plus a backend that creates the session and monitors it.

Is ChatGPT Voice the same as a phone call?

No. ChatGPT Voice runs in the ChatGPT desktop app through your microphone and speakers, and it never touches the telephone network. It is a conversation with the model, not with your prospect.

Can an AI agent receive phone calls instead?

Yes, and this is the easier side in every sense. You route your provider’s SIP trunk to your OpenAI project, your backend accepts the incoming call webhook, and the model handles the conversation. The FCC has stated the artificial-voice prohibition does not extend to technologies used to answer inbound calls, which is a large part of why inbound deployments move faster.

What does it actually cost to let an agent call prospects?

It costs more than the model, and the forgotten line items are the expensive ones. You pay for the API, the SIP provider, the phone numbers, and the engineering time to build and monitor the backend that holds it together. Budget the carrier relationship as a permanent line item rather than a one-time setup cost.

Do I need consent before an AI voice calls a cell phone?

Under 64.1200(a)(1), an artificial or prerecorded voice call to a cellular number needs the prior express consent of the called party absent an emergency purpose, and under (a)(2) telemarketing with that voice needs prior express written consent. The FCC has determined that voice cloning and similar technologies fall under the statutory phrase. Get legal review before you dial.

Should sales teams let an agent run calls alone?

No, not for selling. Use the agent for research, prioritization, drafting, and after-call admin, and keep a rep on the conversation itself. The moment a buyer goes off-script is the moment the deal is won or lost, and that moment needs a person.

Sources

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.

Can Claude Code Make Phone Calls? It Only Runs the Command

TL;DR: Claude Code cannot make phone calls. Anthropic’s tools reference lists every tool it ships with, and they fall into five categories: file operations, search, execution, web, and code intelligence. Nothing in that list dials. The word phone appears exactly once across the whole reference, attached to PushNotification, which sends a desktop notification from a local session and a phone push through Remote Control. A push is not a call. What Claude Code does have is Bash, which runs shell commands and needs your approval except for a built-in set of read-only commands, and MCP, which lets a connected server hand it new tools. So the real answer to can Claude Code make phone calls is that it can run the command that asks a telephony API to dial. With Kixie’s Make A Call API that command is one POST to the app event endpoint with eventname set to call, and it is agent-first by design: Kixie’s documentation says the call begins by ringing the Kixie agent, then connects to the outbound number once that agent picks up. Kixie’s developer API has no endpoint that auto-dials a lead with nobody on the line, and the same Make a Call action is also available as a Zapier step for teams who want the trigger in a workflow tool instead of a script. The rep still does the talking. Claude Code pressed a button. Before you let a script press it, settle three things: which permission rule allows the command, since a don’t-ask-again Bash approval saves permanently per repository into a settings file in your checkout; whether the run is unattended, since a session started with the print flag loads a project’s hooks and MCP servers with no workspace trust dialog and no per-server approval prompt; and where the call record lands, since Kixie’s End Call Webhook returns a call ID, duration, call status, disposition, recording URL and attempt count, and a call your manager cannot inspect is not a process.

Somebody on your team has already asked this. A rep or a technical founder opened a terminal, pointed Claude Code at a repo, watched it run a build and fix a failing test, and thought the obvious next thought. If it can run anything, can it run the dialer? Can Claude Code make phone calls on its own, at night, through a list nobody has time to work?

The question is worth taking seriously, because the answer is not the flat no you get from a disclaimer page, and it is not the breathless yes you get from a demo video either. It is a boundary, and the boundary sits in one specific place. Knowing where it sits is what separates a useful build from a prototype that dials a stranger at two in the morning with nobody accountable for it.

Claude Code has no tool that dials a phone

So what does Claude Code actually ship with? Anthropic publishes a tools reference that lists every built-in tool by name, with a column for whether it needs your approval. It is a long list: Agent, Bash, Edit, Glob, Grep, Read, WebFetch, WebSearch, Write, Monitor, a set of task tools and a set of scheduling tools.

A shallow frosted glass tray holding seven small glass tokens of different shapes, a cube, disc, cylinder, wedge, pebble, pyramid and flat tile, with one empty rounded socket pressed into the tray at the right.

None of them place a call.

Anthropic’s own summary groups the built-in tools into five categories: file operations, search, execution, web, and code intelligence. There is no sixth category for telephony. That is not an oversight anyone is hiding. Claude Code is an agentic coding tool built to read a codebase, change files and run commands, and dialing a phone number is none of those things.

Want a sharper test than a category list? Run a word search across the whole reference. The word phone appears exactly once. It is attached to PushNotification, described as sending a desktop notification from a local session and a phone push with Remote Control or in a cloud session. So the single point where Claude Code’s shipped tool surface touches a phone at all is a notification telling you a job finished.

That is worth sitting with for a second. The tool that reaches a phone is the one that says “I’m done.”

What Claude Code can actually run from the command line

Here is the part people skip, because Anthropic’s own description of what Claude Code can reach is blunt: your terminal, and any command you could run. Build tools, git, package managers, system utilities, scripts. If you can do it from the command line, Claude can too.

So the ceiling is not Claude Code’s ceiling. It is your shell’s ceiling.

So what are you actually asking? Not whether a coding agent has a phone feature. You are asking what happens when a program that can run arbitrary commands is pointed at a telephony API that accepts an HTTP request. The agent does not become a phone, it becomes a caller of an API in the software sense, and the dialing stays wherever it already lived.

Which paths get it there? Two, and they behave differently.

The first is Bash. Claude writes a curl command, Claude Code shows you the command, and you approve it. Per the permission table, Bash commands need approval except for a built-in set of read-only commands, so a request that posts to an outbound calling endpoint will stop and ask you the first time.

The second is MCP. The Model Context Protocol lets Claude Code connect to external tools, databases and APIs through a server, and a connected server hands Claude new tools by name. A telephony vendor with an MCP server gives Claude a tool that looks native. Is it native? No. It is a server running on your machine or at a URL, and Anthropic’s documentation carries a plain warning next to it: verify you trust each server before connecting it, because servers that fetch external content can expose you to prompt injection risk.

Neither path makes Claude Code a dialer. Both make it a thing that can ask a dialer to work.

Voice dictation moves one direction and Claude Code cannot hear a caller

The next assumption worth killing is the audio one. Claude Code does support voice, and that fact gets stretched into something it does not support.

Voice dictation lets you speak your prompt instead of typing it. Your speech streams to Anthropic’s servers and comes back as text in the prompt input. That is the whole feature, and it needs a claude.ai account and a local microphone, which is why Anthropic notes that dictation does not work in cloud sessions or SSH sessions.

Audio goes in. Text comes out. Nothing goes back out as audio.

So what does a live phone call need that dictation does not supply? It needs the other person’s speech converted to something the model can read, a response generated inside a few hundred milliseconds, and that response converted back to audio and pushed onto the call leg while the buyer is still on the line. Dictation supplies one leg of that, into your prompt box, from a microphone sitting next to you.

What about screen control? Claude Code does have computer use, which can open apps, click, type and see your screen, and in principle that could drive a softphone the way a person does. Read the availability line before you build on it. Anthropic describes computer use as a research preview on macOS that requires a Pro, Max or Ultra plan, says it is not available on Team or Enterprise plans, and states that it requires an interactive session, so it is not available in non-interactive mode with the print flag.

Read that last clause again, because it is the one that matters for a sales team. The screen-control path is explicitly unavailable in exactly the mode you would use to run something unattended. And the Team and Enterprise exclusion means the plan tiers most sales organizations buy do not have it at all.

How Claude Code makes a phone call happen

So what does a real one look like? It looks like an HTTP request, and it is less impressive and more useful than the demo videos suggest.

Take Kixie as the concrete case, since Kixie publishes the request shape. Get one constraint straight before the payload, because it shapes everything after it: there is no Kixie endpoint that auto-dials a lead with nobody on the line. The Make A Call API is agent-first by design, and that design is the point rather than a limitation. It takes a POST to the Kixie app event endpoint carrying a business ID, the agent’s email, the target number, a display name, an API key, and an event name of call. That is the entire payload, so any language that can make an HTTPS request can send it, which means Claude Code can write the request and, with one approved Bash command, send it.

Now read what Kixie says happens next, because this is the sentence the whole article turns on. The call begins by calling the Kixie agent. Once the agent picks up the phone call, the call is initiated to the specified outbound number. Kixie names the use case directly: speed-to-lead calls. So the endpoint does not replace a rep, it summons one.

Walk the sequence. A form fills out at 9:14. A script notices. The script posts one request. A rep’s phone rings. The rep answers, and only then does the buyer’s phone start ringing. By 9:15 a human being is on a live call with somebody who raised a hand sixty seconds ago.

That is a genuinely good outcome. So who did the selling? A rep did. The agent compressed the gap between the signal and the conversation, which is the part of speed-to-lead that is a system property rather than a motivational speech. It did not hold the conversation, handle an objection, or read the room when the buyer’s tone changed.

Two more paths are worth naming, and both are quieter. The first keeps the trigger out of your script entirely: the same Make a Call action exists as a Zapier step for Kixie, agent or queue, alongside the PowerList and Cadence actions, which is the better shape when the logic belongs to an operations team rather than an engineer. The second is quieter still. Instead of triggering a call at all, a script adds the contact to a PowerList so the dialing session picks it up in order, using the same endpoint with a different event name. Nobody’s phone rings unexpectedly, because the record simply enters the queue with the rest of the day’s work, which is usually the right answer for anything that is not a fresh inbound signal.

Kixie is explicit about the integration status here, and it is worth repeating rather than softening: Kixie does not have a native Claude integration. Teams connect Claude and Kixie with custom builds on the Kixie API and webhooks, or through Zapier, Make or n8n. If you want the shape of those builds rather than the theory, the Claude integration page and the Claude dialer walkthrough cover the working patterns.

The permission rule decides who can make phone calls with Claude Code

So where does this actually break? Not where people expect. The risk is not that Claude Code dials somebody by accident during a coding session. The risk is what happens to the approval after you click it.

Anthropic is direct about where the control lives. Permission rules are enforced by Claude Code, not by the model. Instructions in your prompt or in a CLAUDE.md file shape what Claude tries to do, but they do not change what Claude Code allows.

Which way does that cut? Both, and the second one is easy to miss. Writing “never call anyone” into a project file is not a control. It is a suggestion to the model. The actual control is the permission rule.

So what does a permission rule actually do when you create one in a hurry? When you answer a Bash prompt with “yes, and don’t ask again,” Claude Code saves that rule permanently, per repository and command, into a settings file at the root of the git repository. It applies to future sessions anywhere in that repository, including sessions started in subdirectories and in worktrees.

What does that mean once a calling command is in it? One tired approval on a Friday writes a rule into a file. That file lives in a repo, and repos get cloned, branched, and handed to the next engineer without anybody reading the permission block on the way through. The question “who is allowed to place outbound calls from this machine” has quietly become a question about a JSON file in a checkout, and nobody in sales operations knows the file exists.

The blunt-instrument answer is worse, because Claude Code has a mode that skips permission prompts entirely and Anthropic attaches a warning to it: use it only in isolated containers or virtual machines where Claude Code cannot cause damage. A machine holding live API keys for your calling platform is not that. If somebody proposes running the dialing script that way, the answer is no, and the reason is in the vendor’s own documentation.

There is a better lever in the same system. For a session that must run without a human at the keyboard, Anthropic documents a setting that denies calls that would otherwise wait on a prompt, so the run keeps moving instead of hanging. Pair that with a narrow allow rule naming the one calling command you intend and nothing else. The rule becomes the policy, and the policy is readable.

An unattended Claude Code run skips the trust prompt

Everything above assumes somebody is watching. What happens when nobody is? The interesting sales use cases are exactly those, so read how the unattended mode actually behaves.

Claude Code runs non-interactively with a print flag and a prompt. That is the form you would put in a cron job or a CI step. Anthropic documents a detail about it that deserves to be on a whiteboard.

Without the bare flag, a non-interactive run loads the hooks in a project’s settings file and connects the servers in its MCP config, even in a folder you have never trusted. That kind of run shows no workspace trust dialog and no per-server approval prompt.

So the mode with the least supervision is also the mode with the fewest confirmation steps. That is not a flaw anybody is hiding, it is a documented tradeoff, and it is exactly the tradeoff a sales engineer should price in before the script goes near a lead list. A repo you cloned to try somebody’s calling example can carry its own hooks and its own MCP server config, and an unattended run will connect them.

Bare mode is the counterweight, because it skips auto-discovery of hooks, skills, custom commands, subagents, installed plugins, MCP servers, auto memory and project instruction files, and it limits Claude to the Bash, file read and file edit tools. Anthropic recommends it for scripts and CI where you need the same result on every machine. For a job whose only purpose is to post one call request, that constraint is a feature.

One more boundary belongs here, because Claude Code ships a sandbox for shell commands that routes network traffic through a local proxy and checks every host against an allowed-domains list that starts empty. Add your telephony host to that list and nothing else, and a compromised script cannot quietly post your lead data somewhere else. The sandbox covers shell commands only, though, so file tools, MCP servers and hooks run outside it, which is another reason to prefer one approved Bash command over a third-party MCP server you did not write.

Where the phone call record lands decides whether this is a process

Here is the test that actually matters, and it has nothing to do with whether the agent can dial.

A frosted glass catch basin holding four translucent glass cards stacked unevenly, with a fifth card caught in mid-air above them, tilted and softly blurred as it falls toward the stack.

Can your manager inspect the call on Monday?

What defines a calling process? Not what initiates it. It is defined by what it leaves behind: the outcome, the recording, the disposition, the attempt history, and whether another rep can pick up the deal when the first one leaves. A script that fires calls and writes nothing back is not a calling process at all, because at the end of the week nobody can say which numbers were tried, how many times, or what the buyer actually said. It is call volume with no evidence attached.

This is where the webhook side earns its keep. Kixie’s End Call Webhook fires every time a call ends and returns the information from that call, including a call ID, the call date, the from and to numbers, duration, call type, call status, disposition, a recording URL, the PowerList the contact came from, and the attempt count with the last attempt timestamp.

Read that field list as a management surface rather than a payload. A deal comes back marked closed lost. Fine. How many real attempts happened before somebody made that decision? Attempt count and last attempt timestamp answer it, and a disposition on its own never does. Call status and duration are what separate a real connected conversation from a dial that rang out into voicemail, and those two fields do more for a coaching session than any volume chart. Disposition is what the rep decided, and the recording URL is what the manager coaches from on Monday morning.

Now put the two halves together, and the shape of the whole thing shows up. The agent posts a request and something rings. The platform posts back a record and something is learned. Call recording and outcome logging is the half that makes the first half worth doing, and it is the half the demo videos never show, because a webhook payload is boring and a ringing phone is not.

If you only build the first half, you have automated the part that was never the bottleneck.

What to inspect before Claude Code touches a live phone call

Treat this as the checklist, not the conclusion. Each item is something you can go look at today.

Start with the permission rule. Open the settings file in the repository where the script lives and read the allow list out loud. If the calling command is approved there under a broad wildcard, narrow it to the exact command. If you cannot explain a rule, remove it.

Then check the run mode. Is the job interactive or does it run with the print flag? If it runs unattended, add the bare flag so it stops loading whatever hooks and MCP servers happen to be in the folder, and pass the config you actually intend.

Then check the key scope. The API key in that script can place calls. Treat it like a key that places calls, not like a build secret, and keep it out of the repository.

Then check the trigger. What event fires the request? A fresh inbound signal justifies ringing a rep immediately. A row in a spreadsheet does not, and that record belongs in a PowerDialer list where it enters the queue in order with everything else.

Then check the record. Point the End Call Webhook somewhere real and confirm the outcome and disposition arrive. Can a manager find yesterday’s calls without asking the person who wrote the script? If not, the record is not a record yet.

Last, who owns it? An agent-triggered calling path has an owner in engineering and an owner in sales operations, or it has neither, and the second case is how a script keeps dialing a week after the person who wrote it changed teams.

Claude Code is a good tool for building this. It is a fine tool for writing the request, the retry logic, the webhook handler and the tests. It is not the dialer, it is not the carrier, and it is not the rep. Keep those four things separate in your head and the build goes fine.

Claude Code phone calls FAQ

Can Claude Code call my phone number directly?

No. Claude Code has no tool that places a call. The closest thing in its tool list is PushNotification, which can send a push to your phone through Remote Control, and that is a notification rather than a call. A call requires a telephony platform, and Claude Code reaches one the same way any script does, by sending a request.

Can Claude Code answer an inbound phone call?

No. Answering a call means holding a phone number, accepting the inbound leg, and exchanging audio in real time. Claude Code does none of those. A platform that owns the number handles the inbound leg, and Claude Code can help you write the handler that reacts to the event afterward.

Can Claude Code run a calling script overnight with nobody watching?

It can run non-interactively with the print flag, which is how most unattended jobs work. Weigh the documented behavior first, because without the bare flag that kind of run connects a project’s hooks and MCP servers with no workspace trust dialog and no per-server approval prompt, so pin the run down with bare mode, a narrow allow rule, and a sandbox domain list before it touches a live number.

Does Claude Code talk to the buyer once the call connects?

No. On the Kixie path, the request rings the Kixie agent first and connects to the outbound number once that agent answers, so a person is on the line from the start. The agent compressed the delay between the signal and the conversation. The rep still runs the call.

Is Claude Code enough to build an outbound calling workflow by itself?

Not by itself. You still need a calling platform with numbers and caller ID handling, a CRM that holds the outcome, a webhook endpoint that survives retries, suppression and consent rules that match the audience, and somebody accountable when a call goes wrong. Claude Code shortens the build. It does not supply the stack.

Sources

  • Anthropic, Claude Code tools reference, for the complete list of built-in tools with their approval requirements, the Bash entry requiring approval except for a built-in set of read-only commands, and the PushNotification entry sending a desktop notification from a local session and a phone push with Remote Control or in a cloud session. https://code.claude.com/docs/en/tools-reference. Reviewed November 15, 2026.
  • Anthropic, How Claude Code works, for the agentic loop, the five built-in tool categories covering file operations, search, execution, web and code intelligence, and the statement that Claude Code can reach your terminal and any command you could run from the command line. https://code.claude.com/docs/en/how-claude-code-works. Reviewed November 15, 2026.
  • Anthropic, Claude Code permissions, for permission rules being enforced by Claude Code rather than the model, prompt and CLAUDE.md instructions not changing what is allowed, a don’t-ask-again Bash approval saving permanently per repository and command to the settings file at the git repository root, the warning to use the prompt-skipping mode only in isolated containers or virtual machines, and the setting that denies calls that would otherwise wait on a prompt. https://code.claude.com/docs/en/permissions. Reviewed November 15, 2026.
  • Anthropic, Run Claude Code programmatically, for the non-interactive print flag, bare mode skipping auto-discovery of hooks, skills, commands, subagents, plugins, MCP servers, auto memory and project instruction files, bare mode limiting access to the Bash, file read and file edit tools, and the note that without bare mode a print-flag run loads a project’s hooks and MCP servers in a folder you have never trusted with no workspace trust dialog and no per-server approval prompt. https://code.claude.com/docs/en/headless. Reviewed November 15, 2026.
  • Anthropic, Let Claude use your computer from the CLI, for computer use being a research preview on macOS, requiring a Pro, Max or Ultra plan, not being available on Team or Enterprise plans, and requiring an interactive session so it is not available in non-interactive mode with the print flag. https://code.claude.com/docs/en/computer-use. Reviewed November 15, 2026.
  • Anthropic, Voice dictation, for speech being transcribed live into the prompt input, the claude.ai account and local microphone requirements, and dictation not working in cloud sessions or SSH sessions. https://code.claude.com/docs/en/voice-dictation. Reviewed November 15, 2026.
  • Anthropic, Connect Claude Code to tools via MCP, for MCP servers giving Claude Code access to external tools, databases and APIs, and the warning to verify you trust each server before connecting it because servers that fetch external content can expose you to prompt injection risk. https://code.claude.com/docs/en/mcp. Reviewed November 15, 2026.
  • Anthropic, Configure the sandboxed Bash tool, for the sandbox covering shell commands only while file tools, MCP servers and hooks run outside it, and network traffic having no direct route out and passing through a local proxy that checks each host against an allowed-domains list that starts empty. https://code.claude.com/docs/en/sandboxing. Reviewed November 15, 2026.
  • Kixie, Make A Call developer documentation, for the POST request to the app event endpoint carrying a business ID, email, target number, display name and API key with an event name of call, and the statement that the call begins by calling the Kixie agent and is initiated to the specified outbound number only once that agent picks up, named as useful for speed-to-lead calls, which is the agent-first behavior this article relies on rather than any auto-dialing endpoint. https://www.kixie.com/developer/make-a-call/. Reviewed November 15, 2026.
  • Kixie, End Call Webhook developer documentation, for the webhook firing every time a call ends and returning the call ID, call date, from and to numbers, duration, call type, call status, disposition, recording URL, PowerList identifier, attempt count and last attempt timestamp. https://www.kixie.com/developer/webhooks-end-call-webhook/. Reviewed November 15, 2026.

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.

Can Claude Send Text Messages? Who Sends Decides the Rules

TL;DR: You can watch Claude send text messages today, and the three paths are not interchangeable. On iOS and Android, Anthropic’s help pages say Claude drafts the message and pre-fills it in your messaging app, you tap send, and Claude does not read existing messages or emails, only creates new content. On macOS there is a real send. Anthropic’s own Read and Send iMessages connector, added in March 2026, exposes send_imessage, read_imessages, get_unread_imessages and search_contacts, drives the Messages app through macOS automation, requires explicit macOS permission prompts, and states plainly that it is not affiliated with or endorsed by Apple. Through the API, Claude still never sends anything itself. Anthropic documents that client tools run in your application, that Claude responds with stop_reason tool_use, and that your code runs the operation and sends back a tool_result. That split is the whole buying decision, because CTIA’s Messaging Principles and Best Practices classify traffic by who the sender is rather than by which app sent it. If the sender is a business, an organization, or an individual acting on behalf of one, the messaging is Non-Consumer, and CTIA states that automation used in whole or in part by a Non-Consumer to facilitate messaging is not typical Consumer operation. The Consumer attributes CTIA publishes run to 15 to 60 messages a minute, 1,000 a day, one telephone number per Consumer, 100 recipients per message, a one to one ratio of outgoing to incoming, and 25 repetitive messages, and a footnote says that clearing those numbers still does not make a business sender a Consumer. So pick the number before you pick the integration. A personal thread has nowhere to put the seven opt-in details CTIA tells senders to retain, and no way to produce them a year later.

Claude Can Send Text Messages Three Different Ways

Somebody on your team has already tried it. They asked Claude for a follow-up text after a rough call, liked what came back, pasted it into their phone and sent it, and then asked the obvious next question, which is whether the whole desk could work that way. That is where this stops being a question about the model. It becomes a question about your phone system.

There are three live paths and they differ in one specific way. Not quality. Not speed. What separates them is which telephone number the message leaves from and who owns that number, which sounds like a detail and turns out to decide the opt-out, the record, the deliverability and whether a second rep can ever pick up the thread.

  • The mobile apps. Claude writes the message and hands it to your phone’s messaging app. Your thumb sends it, from your personal number.
  • The macOS connector. Claude operates the Messages app on your Mac and sends without you touching the keyboard, still from your personal Apple identity.
  • The API or an automation layer. Claude decides and drafts. Separate code you control sends the message from a business number that is registered to send it.

Those are three different senders wearing one familiar name. Everything downstream, the opt-out, the record, the deliverability, the thing you hand a reviewer eighteen months from now, follows from which one a rep used. Work the paths in order.

On Your Phone, Claude Drafts the Text Message and You Send It

Start with the version most reps meet first. Anthropic’s help article for iOS says Claude can draft and send messages through the Messages app or any messaging platform such as WhatsApp, Slack or Messenger. Read the steps underneath and the word send does a lot of quiet work. You ask Claude to draft. Claude prepares a preview. You tap the message card, which opens Messages with the content already filled in. You review it, change what you want, and send it the normal way.

The Android article describes the same shape for your default messaging app and third-party apps. Both pages say no permission is needed, because the hand-off runs on the device’s built-in sharing system. Both pages also carry the same boundary. Claude does not read existing messages or emails, only creates new content.

So what did you actually buy here? A faster first draft. Nothing else moved.

That is worth saying out loud, because the demo is genuinely good and the conclusion people draw from it is wrong. The rep still opens the thread, still reads what Claude wrote, still decides whether the customer will take it that way, and still presses send, which means every constraint that was already on the rep is still on the rep. There is no queue, no list, no schedule, no retry, no suppression check. One rep, one message, one thumb. If the pitch inside your company is that this frees the team from texting, that is not what the demo showed and it is not what the documentation says.

Useful, though, and worth keeping for what it is. A rep who stalls on wording after a rough call, opens a thread, types two words, deletes them and goes to get coffee instead now has a draft in front of them in seconds, which removes the pause rather than the work. That is a real fix for a real behavior and it costs nothing to allow.

Watch Claude Send Text Messages From a Mac

Now the part that changes the question, because this is where the demo stops being a demo.

Anthropic publishes a connector called Read and Send iMessages, made by Anthropic and added in March 2026. It exposes four tools: send_imessage, search_contacts, read_imessages and get_unread_imessages. It works by using macOS automation features to interact with the Messages app, and it requires explicit permission through macOS security prompts before it touches Messages or Contacts. The listing also says it is not affiliated with or endorsed by Apple.

Read that tool list again. Send, search contacts, read, and read unread. What does that combination give you that the phone apps do not? A loop. Claude can look up a contact, send, read what came back and send again without a person in the middle, and that is a different product from the one on your phone.

Two details in that description are load-bearing and most write-ups skip both.

First, it drives the Messages app rather than calling a messaging API, so whatever identity that Mac is signed in with is the identity on every message it sends, and whatever the Messages app can do on that one machine is the hard ceiling on the whole arrangement.

Second, Anthropic states that it is not affiliated with or endorsed by Apple. An automation that operates a consumer app is not the same thing as a supported business messaging channel, and the connector page does not pretend otherwise, which is more candour than most write-ups about it manage.

Here is the failure mode, and notice that it is not a legal abstraction. Say a rep connects it on a Friday, texts nine customers from their own number over the weekend and books two meetings, which is a good weekend by any standard. Monday the rep is out. Three of those nine reply. Nobody else can see the thread, because the thread is on a laptop in somebody’s kitchen, which means the manager cannot read it, the CRM has no record that it happened, and the next rep to touch those accounts starts from nothing. That is not a compliance problem yet. That is a hole in the pipeline.

Who Sends the Text Message Decides Which Rules Apply

The compliance problem arrives next, and it does not care which laptop was used.

A translucent glass channel passing through an opening in a thick glass plate, smooth on one side of the plate and finely ribbed on the other, on a pale violet ground

CTIA publishes the Messaging Principles and Best Practices, the document the wireless carriers work from. It sorts all messaging into two buckets, Consumer and Non-Consumer, and the executive summary states the test in one sentence. If the Message Sender is a business, organization, or other entity, or an agent, representative, or other individual acting on behalf of a business, organization, or other entity, the messaging is considered Non-Consumer.

Read the clause about the individual again, because that is the half most teams skip. A rep texting a customer about a renewal is an individual acting on behalf of a business, which is the exact wording, so the traffic is Non-Consumer before anyone asks what it was sent from. The handset does not change that. The Apple ID does not change that. The classification follows the sender and not the device.

CTIA then addresses automation directly, which is unusual and lucky for anyone trying to answer this question. It grants the sensible case first. Some Consumers use automation to help them respond, and it gives the example of an auto-reply that tells a caller the person is busy or driving. That kind of help with composing and sending falls inside typical Consumer operation. Then comes the boundary. Automation used in whole or in part by a Non-Consumer to facilitate messaging is not typical Consumer operation.

That sentence was written before this connector existed and it lands on it exactly. Which case are you in? An assistant that drafts your reply to a friend is the auto-reply case and nobody minds. The same assistant working a customer list on behalf of your employer is the other case, and CTIA draws that line in terms rather than leaving it to interpretation.

One practical note on scope. Those principles are voluntary industry best practices, not a statute, and they say so. They are also what the carriers who deliver your messages use, which makes them the operating rules whatever their legal weight. For the statutory layer, read the four separate rulebooks that govern a marketing text rather than treating carrier policy as law.

The Consumer Thresholds a Sales Desk Blows Past

CTIA also publishes what typical Consumer operation looks like in numbers. These are the attributes it lists.

  • Throughput of 15 to 60 messages per minute. A Consumer typically cannot originate or receive more than about one message per second.
  • Volume of 1,000 per day. Only in unusual cases do Consumers send or receive more than a few hundred in a day, and a Consumer cannot typically send or receive continuously over a long period.
  • One telephone number assigned to or used by a single Consumer.
  • 100 distinct recipients per message. A Consumer typically sends to a limited number of recipients, with ten given as the example.
  • A one to one ratio of outgoing to incoming messages per telephone number, with some latitude either way, because Consumer messages are conversational.
  • 25 repetitive messages. Typical Consumer behavior is not to send essentially or substantially repetitive messages.

Look at the balance line and the repetition line together, because those are the two a sales desk breaks first and neither of them is about volume. A rep working a follow-up list sends many messages and hears back from a few, which is the exact opposite of the one to one ratio CTIA describes as conversational. The same rep sends a near-identical message to forty people, which is the repetition attribute. How much traffic does that take? A quiet Tuesday does it.

Now the part that stops the obvious workaround. CTIA attaches a footnote saying that even where traffic satisfies the throughput, volume and balance numbers consistent with typical consumer operation, it still qualifies as Non-Consumer messaging because it is sent by a Non-Consumer. So throttling does not buy reclassification. Staying under 1,000 a day does not make a business sender a Consumer. The attributes are illustrative and the sender is determinative.

This is the whole reason to read the thresholds at all, and it is the opposite of how most people read them. They are not a speed limit you can drive under. They describe someone you are not.

What a Personal Thread Cannot Record About a Text Message

Set the classification aside for a second and ask the operational question instead. What has to exist on the day somebody asks why a particular contact got a particular text?

Two shallow frosted glass trays on a pale violet ground, the left one empty and the right one holding a loose fan of thin translucent cards

CTIA lists what Message Senders should retain to document opt-in consent, where applicable. The timestamp the consent was acquired. The medium it came through, such as a web form, a paper sign-up or a text keyword. A capture of the experience, meaning the language and the action used to secure it. The specific campaign the opt-in was for. The IP address used to grant it. The Consumer phone number it covers. And the identity of the individual who consented.

Seven items. Where do any of them live in an iMessage thread?

There is no field for the timestamp of consent, because a thread only knows when a message was sent and has no opinion about when permission was given. There is no campaign, because a personal thread has no campaigns. There is no capture of the opt-in language, because the opt-in happened somewhere else entirely and nothing carried it across. The thread holds exactly one of the seven, the phone number, and it holds that one by accident.

CTIA also says an opt-in should not be transferable or assignable and should apply only to the campaigns and the specific Message Sender it was obtained for. A customer who agreed to appointment reminders from your company agreed to those, from that sender. A text from a rep’s personal number is a different sender making a different ask. The registration and delivery mechanics behind that rule are worth reading separately, because they are where most teams first notice something is wrong.

And notice how this failure presents itself, because it does not present as an error. The rep’s tool reports that the message went, the customer may well reply, and every dashboard anyone looks at stays green for as long as nobody asks a specific question. Nothing surfaces until the day somebody asks for the record, and on that day the answer is that there is no record, there is a laptop.

What does that cost you in an ordinary week, before any of the serious scenarios? A rep leaves and takes nine live conversations with them. A manager coaching a struggling rep can read every call and none of the texts, which is half a diagnosis. Two reps text the same customer inside the same hour because neither of them can see the other’s thread. The opt-out somebody sent on Saturday afternoon sits unread in one personal inbox while the rest of the team keeps messaging that contact on Monday.

Build a Claude Text Message Path That Leaves a Record

So build the other path. It is not harder, it is just explicit.

Start from how the Claude API actually works, because it already splits the job the way you need it split. Anthropic’s tool use documentation says client tools run in your application. Claude responds with stop_reason tool_use and one or more tool_use blocks. Your code runs the operation and sends back a tool_result.

Translate that out of the documentation and onto the desk. Claude never sends the text message. Claude asks for a text message to be sent, and something you wrote decides whether to honor the request. That is not a limitation you are working around. That is the seam you want.

So where should the opt-out check live? Not in the prompt, where it is a probability that a model remembered an instruction, but in the few lines of code sitting between the tool_use block and the send, where it is a yes or a no and you can read it in a log afterwards. Same for the business-hours rule, the per-contact frequency cap, and the check that this contact actually opted in to this campaign. A model that is right almost every time is the wrong place for a rule that has to be right every time.

On the Kixie side, the connection is a custom build and the Claude integration page says exactly that. Kixie does not have a native Claude integration. Teams connect Claude and Kixie with custom builds on the Kixie API and webhooks, or through Zapier, Make or n8n. Never let anyone describe that as one-click or built in.

The outbound half is one documented request. Kixie’s Send SMS automation takes a POST to the app event endpoint with a business ID, a target number, the message, and an API key, and it sends from the business’s own numbers. The documentation puts it on the Professional billing tier or higher as of November 15, 2026. Because the key travels in the request, put a small server between Claude and Kixie rather than placing the key inside a tool definition a user can read.

The inbound half is what makes it a conversation instead of a broadcast. Kixie’s webhooks post call and message events to a URL you own, so a reply can reach your code, your code can update the CRM, and Claude can be handed the thread as context for the next draft. Kixie’s webhooks and API surface is where that side gets configured.

What changes once the send runs through a business number? The message becomes part of a campaign registered for that purpose, the reply lands in a shared inbox that a second rep can answer, and the thread is logged against the contact in the CRM so the manager can read it and the next rep inherits it instead of starting cold. A stop request hits a suppression list that every connected tool reads, rather than one phone that happens to be on aeroplane mode. Kixie’s business texting feature page covers the A2P 10DLC side of that, which is the registration layer the carriers check.

None of that is Claude’s job and none of it should be. Claude is good at the sentence. The number is good at the record. Keep them in their lanes.

What to Check Before You Let Claude Send Text Messages

Before anyone connects anything, settle these. They take an afternoon and they decide everything after it.

  • Which number does the message leave from? Answer this first. If the answer is a rep’s personal number, stop and finish the other questions anyway, because the answers will change your mind.
  • Who is the sender, in CTIA’s terms? If a person is texting on behalf of the company, the traffic is Non-Consumer whatever handset is involved.
  • Where is the opt-in record? Name the system and the field. If you cannot name both, you do not have one.
  • What happens when somebody replies stop? Trace it. The suppression has to be enforced by the system that sends, not by the model that drafts, and every connected tool has to read the same list.
  • Who else can see the thread tomorrow? If the answer is one rep and one laptop, the conversation is not in your pipeline.
  • What does the rep no longer do by hand? If the honest answer is only typing, you have a drafting aid, and you should buy it as one.

Then run it on a small list of people who agreed to be on it and read every message that goes out for the first week. Not because the drafts will be bad. Because the first time your rules and your workflow disagree with each other, and they will, you want to find that out on ten contacts rather than on four hundred.

The useful version of this is boring, and that is the point. Claude reads the call summary, drafts the follow-up and proposes it, your code checks the suppression list and the campaign before anything leaves, the business number sends it, and the reply comes back through a webhook into a shared inbox where somebody other than the original rep can answer it. Nobody is impressed by the demo. The record survives the rep.

For a worked build of the same shape on the other major model, the ChatGPT SMS bot teardown walks the costs and the consent handling end to end.

Claude Text Message FAQs

Can Claude send a text message on its own?

Not through the API. Anthropic documents that client tools run in your application, that Claude replies with stop_reason tool_use, and that your code runs the operation. On macOS the Read and Send iMessages connector does send, but it does so by operating the Messages app on your Mac after you grant macOS permission. On iOS and Android, Claude drafts and pre-fills and you press send.

Does the macOS iMessage connector work for business texting?

No, not as your business texting channel, even though it genuinely does send real messages. The classification is the problem. CTIA treats messaging sent by an individual acting on behalf of a business as Non-Consumer, and says automation used by a Non-Consumer to facilitate messaging is not typical Consumer operation. There is also nowhere in a personal thread to keep the opt-in details CTIA tells senders to retain.

Can Claude read the replies to a text message?

Only on macOS, and in a build you wire yourself. On iOS and Android the answer is no, and both help pages say so, because Claude does not read existing messages or emails and only creates new content. The macOS connector exposes read_imessages and get_unread_imessages, so it can. In a custom build, replies reach Claude only if you route them there, usually through an inbound message webhook.

Do we still need 10DLC registration if Claude writes the texts?

Yes, and the question has the dependency backwards. Registration attaches to the number and the campaign, not to whoever or whatever wrote the words. A model drafting the copy changes nothing about what the carriers check.

Is iMessage the same as SMS for business messaging rules?

Treat them as different channels with the same obligation. An iMessage between two Apple devices does not travel the same path as an SMS, and the macOS connector drives the consumer Messages app rather than a business messaging service. The sender test that puts you in Non-Consumer territory does not turn on the transport.

What is the smallest useful version of this?

Let reps use the mobile app to draft, and send everything outbound from the business number through your normal texting tool. You get the speed benefit on day one and you change nothing about the record. Add the API path later, when you have a rule you want enforced in code.

Sources

  • Anthropic, Read and Send iMessages connector listing, for the connector being made by Anthropic, added March 2026, the send_imessage, search_contacts, read_imessages and get_unread_imessages tools, the use of macOS automation features to interact with the Messages app, the explicit macOS security permission prompts, and the statement that it is not affiliated with or endorsed by Apple Inc. https://claude.com/marketplace/connectors/read-and-send-imessages. Reviewed November 15, 2026.
  • Anthropic Help Center, Use Claude with iOS apps, for drafting and sending through the Messages app or platforms such as WhatsApp, Slack and Messenger, the review and tap-to-send steps, the use of the device built-in sharing system, and the statement that Claude does not read existing messages or emails and only creates new content. https://support.claude.com/en/articles/11869619-use-claude-with-ios-apps. Reviewed November 15, 2026.
  • Anthropic Help Center, Use Claude with Android apps, for the same draft and hand-off behavior on the default messaging app and third-party apps, the note that no permissions are required because it uses the Android sharing system, and the same limit on reading existing messages. https://support.claude.com/en/articles/11869629-use-claude-with-android-apps. Reviewed November 15, 2026.
  • Anthropic, Tool use with Claude, for client tools running in your application, Claude responding with stop_reason tool_use and one or more tool_use blocks, and your code running the operation and sending back a tool_result. https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview. Reviewed November 15, 2026.
  • CTIA, Messaging Principles and Best Practices, May 2023, section 4 on Consumer and Non-Consumer traffic classification, for the sender test covering a business, organization or other entity or an individual acting on behalf of one, the Consumer messaging automation clause, Exhibit I attributes of typical Consumer operation at 15 to 60 messages per minute, 1,000 per day, one telephone number, 100 distinct recipients, a one to one outgoing to incoming ratio and 25 repetitive messages, the footnote that meeting those numbers does not reclassify a Non-Consumer sender, the opt-in documentation items in section 5, and the rule that an opt-in is not transferable or assignable. https://api.ctia.org/wp-content/uploads/2023/05/230523-CTIA-Messaging-Principles-and-Best-Practices-FINAL.pdf. Reviewed November 15, 2026.
  • Kixie, Send SMS developer documentation, for the POST request to the app event endpoint with businessid, target, message and apikey, the sms event name, and the Professional billing tier requirement. https://www.kixie.com/developer/automation-send-sms/. Reviewed November 15, 2026.

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.

Best Dialer for Solar Sales Starts With the 2026 Pitch

TL;DR: The best dialer for solar sales is a script-control problem now, because the number the old pitch leaned on is gone. The IRS states the Residential Clean Energy Credit is not available for any property placed in service after December 31, 2025, and Public Law 119-21 terminated it at section 70506. A setter who still opens with a 30% federal credit for a homeowner buying the system outright is misstating total cost, and the FTC telemarketing rule treats misstated total cost as a deceptive practice. One deadline is still running. Section 48E, the clean electricity investment credit, stops applying to solar property placed in service after December 31, 2027, with energy storage carved out of that cutoff, so a lease or power purchase pitch is measured against an installation date rather than a signature date. That changes what you are buying. You need a dialer that pushes a script change across the whole floor the same day, records what was actually said, writes utility, bill size, ownership and financing path back to the CRM during the call, and re-works an aged list without replaying a dead pitch. On the Kixie side, PowerDialer runs one line on the Single-Line plan and up to 4 lines on Multi-Line, and call recording, Voicemail Drop, live call coaching, Team SMS, rule-based routing and bi-directional CRM integration sit on all three plans. ConnectionBoost, Conversation Intelligence, DNC and AI Human Voice Detection are add-ons, not defaults.

Most buying guides for this category never mention solar. They rank dialers by lines, price tiers and integrations. A solar director reads that list and buys on dial volume. Volume is not the constraint on a solar floor right now. The constraint is that the sentence your setters have said ten thousand times stopped being true on January 1. The floor is still saying it.

So the question is not which dialer is fastest. The question is which dialer lets you change what a hundred reps say by Tuesday, and then prove on Friday that they said it.

Why the best dialer for solar sales is a script problem now

Start with the fact, because everything downstream depends on it. The IRS page for the Residential Clean Energy Credit states that the credit equals 30% of the costs of new, qualified clean energy property for a home installed anytime from 2022 through December 31, 2025. Then the same page states the limit. The credit is not available for any property placed in service after December 31, 2025. The statute behind it is Public Law 119-21. At section 70506 it struck the old 2034 end date out of the credit and replaced it with a cutoff for expenditures made after December 31, 2025. That is the whole change.

Read what that does to a call. A homeowner who buys the system with cash or a loan in 2026 does not get the 30% back. The rebate line, the payback math, the monthly comparison against the utility bill, the urgency close built on a credit stepping down, all of it was built on a number that no longer exists for that buyer.

Now the compliance part, stated narrowly. The FTC telemarketing rule makes it a deceptive practice to misrepresent the total costs to purchase, receive, or use the goods or services in a sales offer. It says directly or by implication, which covers an implied net price. A second clause covers any material aspect of the performance, efficacy, nature, or central characteristics of what is being sold. So where does a quoted net cost land? A net cost that subtracts a credit the buyer cannot claim is a misstated total cost. It does not matter that the rep believed it. It does not matter that the slide deck still says it.

A dialer demo will never raise this. It is the part that decides the purchase. A solar floor is a script distribution system before it is anything else. So the best dialer for solar sales is the one that gets a corrected sentence into every rep’s mouth and leaves evidence of which version they used. Nothing else on the feature list outranks that this year.

So the evaluation question becomes concrete. How long does it take you to change the opener for every rep on every list, and how do you confirm it took? If the answer is a chat message and a manager walking the floor, that is not a control. That is a hope.

  • Where the live script actually lives, in the dialer or in a document somebody has to open
  • Whether a manager can hear a call while it is happening and say something only the rep hears
  • Whether calls are recorded by default on every plan, or only on the tier you did not buy
  • Whether you can pull ten calls from Thursday afternoon and tell which script version the rep used
  • How many systems you have to edit when the pitch changes again

Kixie puts live call coaching and call recording on all three plans rather than behind a tier, with included cloud storage for the recordings. So the listen-and-correct loop does not depend on which plan a solar team ended up on. Conversation Intelligence, the add-on that analyzes those calls, is a separate premium line item. Worth knowing before you build a QA process that assumes it.

The solar deadline still running is an installation date

The 30% credit for a homeowner-owned system is finished as of property placed in service after December 31, 2025. A different credit is not, and the difference matters on the phone.

A long violet glass rail on a pale ground carrying three pale glass spheres spaced at widening intervals, with a small upright glass gate set across the rail and a bare stretch of rail at the far end.

Public Law 119-21 also rewrote section 48E, the clean electricity investment credit. At section 70513 it added a termination for wind and solar facilities. The text is specific. The section does not apply to qualified property placed in service by the taxpayer after December 31, 2027 that is part of an applicable facility. An applicable facility includes one that uses solar energy to produce electricity. There is one carve-out written straight into the paragraph. It does not apply to energy storage technology placed in service at an applicable facility.

Three operational consequences, in order.

First, the credit follows ownership. Under a lease or a power purchase agreement the homeowner does not own the system, so a credit on that system is not theirs to claim in the first place. That is a question for the homeowner’s accountant and it does not belong in a closing line on a first call.

Second, the clock is an installation clock. The statutory test is property placed in service, not a contract signed. On a floor booking appointments in November for installs that land after permitting, interconnection and a utility queue, a deal signed late in the window may not be a deal placed in service inside it. Your pipeline math has to count backwards from the install date, and your setters should not be promising a cutoff they cannot control.

Third, storage is treated differently from the panels. The exception for energy storage technology at an applicable facility is in the rule text, so a battery attach is not on the same 2027 clock the solar property is on. That is a product fact worth knowing before a rep improvises one.

None of this makes a dialer. It makes a dialer requirement. Can you split the list by financing path before the first dial? You have to be able to. A cash buyer and a lease prospect are now two different conversations with two different true statements in them. One blended script gets one of them wrong, every time.

Solar dialing modes, and what each one costs the floor

Vendors blur these categories on purpose. Define them by what the rep actually does, not by the label on the pricing page.

Click to call

The rep picks a record and clicks. One call at a time, full control, no list advance. Fine for a closer calling back twelve warm appointments. On a setter floor it is the slowest thing in the building. The dead time between calls is where call hesitancy lives. What does a rep do after three rejections in a row? Finds reasons to read notes.

Single line power dialing

The rep loads a list and the system advances it. One number at a time, the rep hears every ring, and nothing gets dialed that the rep is not present for. This is the mode most solar setter floors should start in, because the rep is on the line from the first word and nothing arrives as a delayed connection. Kixie’s PowerDialer runs single line on the Single-Line plan and up to 4 lines on Multi-Line.

Multi-line parallel dialing

The system dials several numbers at once and connects the rep to whoever picks up first. Throughput goes up. Two costs go up with it. The first is that somebody who answered gets dropped or hears dead air while the rep is already on another call, and that is the abandoned-call problem the FTC telemarketing rule writes a specific limit around. The second is subtler and specific to your situation. Parallel dialing multiplies whatever script is loaded, so a floor running the wrong opener at four lines is wrong four times as fast.

Kixie caps PowerDial at up to 4 lines. Several products in this category advertise more. Treat a high line count as a number you have to earn with answer-rate data and a clean list, not as a feature you buy and switch on.

Predictive dialing

The system predicts rep availability and dials ahead of it. This is where abandonment rates and dead-air complaints come from on consumer lists, and consumer residential is exactly what a solar list is. Most solar teams do not need it and should not start there.

What to actually pick

Setters on cold residential lists usually want single line first, then a controlled move to parallel once answer rate, list quality and script version are all stable. Closers running scheduled appointments want click to call and nothing else. If one product cannot do both without a second license, you are going to end up with two systems and two sets of call records.

What a solar dialer has to write back after every call

A solar appointment is worthless to the closer without four or five facts attached. A setter who collects them on a notepad has not collected them. So the question is not whether the dialer integrates with your CRM. Which fields does it write, when, and did the rep have to leave the call to do it? That is the question.

A tall violet glass plate with six slotted recesses, five holding pale glass chips and one slot left empty, with a single loose chip lying on the pale ground beside it.

The qualification set is specific to this business.

  • Homeowner or renter, because a renter is a disqualification and the fastest disqualification is the most valuable minute in the day
  • The utility, because rate structure, net metering rules and the interconnection queue are utility-specific rather than state-specific
  • Monthly bill size, because it is the number that makes the pitch concrete and the one the homeowner can read straight off a statement
  • Roof age, shading and condition, because an install that needs a roof first is a different timeline and a different close
  • Financing path, cash, loan, lease or power purchase agreement, because after the credit change these are no longer the same conversation
  • Disposition and next step, including who owns the follow-up and when it fires

Then the part most teams skip. A disposition is not a next step. Not interested closes a record. Call back after the next bill lands creates work. That work has to show up in somebody’s queue without a human remembering it. If a rep has to open a second tab to schedule the callback, the callback gets scheduled sometimes.

Kixie writes call activity back through its bi-directional CRM integration on all three plans, and publishes integrations with HubSpot, Salesforce, Zoho and Pipedrive. It does not publish an integration with a solar-specific proposal or design platform. So if your closers live in a solar CRM, map that handoff before you sign anything. The usual answer is a webhook or an automation layer. The usual failure is finding that out in week three.

Working an aged solar list without re-running a dead pitch

Every solar company is sitting on a list of people who took a call in 2024 and did not buy. That list is the cheapest pipeline in the building. It is also the most dangerous one right now. Why did those prospects pass the first time? Usually payback math, and the payback math has since moved in the wrong direction.

Re-working it is a real motion. Replaying the old pitch on it is not.

A workable re-touch pattern looks like this. Segment the aged list by financing path and by why the deal died. Write one honest opener per segment that does not reference a credit the person cannot claim. Run it in a small block first, listen to twenty calls, and only then put it in front of the whole floor. Voicemail Drop lets a rep leave a recorded message and move to the next number instead of repeating the same sixty seconds by hand, and it is on all three Kixie plans. Those messages are recorded by a rep, which is the point here, because a message recorded last year says last year’s pitch.

Texting carries the appointment confirm rather than the pitch. Team SMS and SMS templates are on all plans. Keep the template library versioned the same way you version the script, because a template folder is exactly where an obsolete claim survives longest.

Consent and do-not-call rules apply to this list the same way they apply to a new one, and an aged record is not a consent record. The federal telemarketing rules on calling windows, registry scrubbing and consent live in 47 CFR 64.1200, and Kixie offers a DNC compliance add-on rather than including registry scrubbing by default. Decide who owns the scrub before you reload the list, not after.

If the old opener is the thing you are replacing, it is worth rereading how a solar phone script is actually built and rewriting from the structure rather than patching the one sentence that broke.

Caller ID is a solar problem before it is a dialer problem

Solar is one of the most-dialed consumer categories in the country. Your numbers burn fast. Your answer rate falls for reasons that have nothing to do with your script. A dialer can help here. It cannot fix it.

Local presence dialing matches the outbound number to the prospect’s area. ConnectionBoost is Kixie’s premium add-on aimed at connection quality and dropped calls. Both are tools against a specific failure, not a general upgrade. The underlying work is number hygiene and call volume per number. Stop hammering the same list with the same number at the same hour every day. The connection-rate side of this for solar teams is covered in more depth in a dedicated look at local presence dialing on solar lists.

Demo questions that expose a solar dialer gap

Ask these on the call, not in the trial. A vendor who has to go check will tell you more than the answer does.

  • Show me changing the script for every rep on one list, right now, and then show me where I see that it changed
  • Pull a call from last Tuesday afternoon and play the first thirty seconds. How many clicks was that
  • A setter books an appointment. Show me every field that lands in the CRM without the rep typing it
  • What happens to the person who answered when my rep is already connected on another line
  • Which plan includes call recording, and what does it cost to keep the recordings
  • Can I run one team on single line and another on parallel under the same account
  • What is the line count ceiling, and what happens to answer rate at that ceiling on a residential list
  • Who scrubs against the do-not-call registry, you or us, and is it an add-on
  • Show me the aged-list workflow, including the recorded voicemail and the text confirm

Two answers should stop the conversation. If nobody can show you a script change propagating, you are buying dial volume and managing copy by rumor. If recording sits on a paid tier and you are running a consumer residential list in a vertical with live misrepresentation exposure, the cheap plan is not cheap.

What to inspect in the first two weeks of a solar dialer pilot

Pick one team, one list segment, and one script version. Do not pilot four things at once. When the number moves, which thing moved it? You will not know.

  • Median time from lead assignment to first live conversation, and the outliers separately from the median
  • Answer rate per outbound number, not just per campaign, so you can see a number degrade before it dies
  • Appointments booked per connected conversation, which is the script metric, against dials per hour, which is the dialer metric
  • Share of booked appointments arriving with utility, bill size, roof note and financing path filled in
  • Closer-reported no-show rate, split by whether a text confirm went out
  • How many calls a manager actually listened to, and whether coaching changed a named behavior the following week

That last one is the one teams skip and the one that pays. Dials per hour is easy to move and easy to fake. Appointments per connected conversation only moves when what the rep says changes, which is the whole reason this purchase matters this year.

Solar sales dialer questions

Is a power dialer or a predictive dialer better for solar sales

Start with a single line power dialer. A solar list is consumer residential, the rep needs to be present from the first word, and predictive dialing introduces abandonment exposure and dead air on exactly the kind of list that generates complaints. Move to parallel dialing only after answer rate and script version are stable, and treat the line count as something you raise on evidence.

Does the solar pitch really have to change in 2026

For a homeowner buying the system, yes. The IRS states the Residential Clean Energy Credit is not available for any property placed in service after December 31, 2025. A quoted net price that still subtracts 30% for that buyer misstates total cost, and the FTC telemarketing rule treats misstated total cost in a sales offer as a deceptive practice. Change the opener, change the templates, and keep the recordings.

Should a solar dialer include texting

Yes, for confirms and reschedules rather than pitching. A text confirm before an appointment is the cheapest no-show reduction available to a setter floor. Keep the templates under the same version control as the script, because an old template is the most common place an obsolete claim survives. Kixie cadences do not send email, so an email confirm is a separate system.

Sources

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.

Best Dialer for Mortgage Brokers Starts With Licensing

TL;DR: The best dialer for mortgage brokers is the one that knows which states each loan officer is licensed in before it connects the call. Regulation H makes a state prohibit an individual from engaging in the business of a loan originator with respect to any dwelling or residential real estate in that state unless the individual first registers through the NMLSR, obtains a unique identifier, and obtains and maintains a valid loan originator license from that state. The license follows the property. It does not follow the rep’s desk. Regulation Z puts the duty on the firm rather than the individual, because the loan originator organization has to ensure every individual originator is licensed or registered before that person acts as a loan originator on a dwelling-secured transaction, and the organization is the party that configures routing. That turns an ordinary dialer setting into a compliance control. Exposure also attaches earlier than most buyers assume. Regulation H counts representing to the public through advertising or other means of communicating that you can take an application or offer terms, counts communicating directly or indirectly to reach a mutual understanding about prospective loan terms, and requires states to be able to impose civil money penalties on individuals who represent themselves as loan originators without a valid license. Local presence caller ID does not solve any of this. An area code is not a license, and the FCC caller identification rules let a telemarketer substitute only the seller’s name and the seller’s customer service number, which must permit any individual to make a do-not-call request during regular business hours. Build the shortlist around four things a demo can prove in an hour. Routing that reads a license field. Dialing modes that show the property state before the call connects. Message templates that carry the right identification per state. And a call record an examiner could read, because Regulation H gives state supervisory authorities power to examine any books, papers, records or other data of any loan originator operating in the state. Kixie’s rule-based routing, call recording with included cloud storage and bi-directional CRM integration are on all three plans; the PowerDialer runs one line on the Single-Line plan and up to four lines on the Multi-Line plan.

What a mortgage broker dialer has to know before it connects a call

Take a hypothetical floor. A refinance lead comes in late in the afternoon, the property is in a state where only half your loan officers hold a license, and the queue hands the lead to whoever has been idle longest. The dialer does not know which half is which, because nobody ever told it.

That is the whole problem in one sentence. Most dialer buying guides for this market compare line counts, pricing tiers and CRM connectors. Those matter. They are not the first question for a licensed business. So what is? Which rep is allowed to have this conversation at all.

Start with where the license lives. The Bureau of Consumer Financial Protection’s Regulation H says a state, to run a compliant program, must prohibit an individual from engaging in the business of a loan originator with respect to any dwelling or residential real estate in the state unless that individual first registers as a loan originator through the NMLSR and obtains a unique identifier, and obtains and maintains a valid loan originator license from the state. Read the words that do the work: in the state. The test is anchored to the dwelling. Not to where the rep sits, not to where the company is headquartered, and not to the area code on the screen.

So where does the lead’s property state live in your stack right now? For most brokerages the honest answer is a form field nobody reads until underwriting. That is too late. It is the first field your dialer needs, and it needs it before the connection, not in the disposition notes afterward.

Now look at who carries the duty. Regulation Z puts it on the company. A loan originator organization that is not a government agency or state housing finance agency must ensure that each individual loan originator who works for it is licensed or registered to the extent required, before the individual acts as a loan originator in a consumer credit transaction secured by a dwelling. That is an operational obligation on the firm. The firm is the party that writes the routing rules, builds the calling lists, and decides what the template says.

Which means the dialer configuration is not an IT preference. It is one of the places that duty gets discharged or dropped. Who owns that setting at your shop? If the answer is whoever set up the account two years ago, start there.

Why dialer routing for a mortgage broker is really a licensing rule

Ask a vendor how their routing works and you will hear three answers: round robin, longest idle, and skills based. All three are fine. None of them knows anything about licensing unless you give it a field to read.

Pale lilac and pink scene in frosted glass: five small glass discs drift from the left toward an upright glass plate cut with one narrow slot, three discs stopped flat against the plate face and one passing cleanly through the slot.

Walk the queue. The lead lands. The automation fires. The system picks the next available rep by whatever rule you set. The rep answers, says their name, and starts asking about the current rate and the balance. Where in that sequence did anything check a license? Nowhere. That is the gap.

Here is why that matters earlier than people expect. Regulation H says an individual engages in the business of a loan originator if, in a commercial context and habitually or repeatedly, the individual represents to the public, through advertising or other means of communicating or providing information, that the individual can or will take a residential mortgage loan application and offer or negotiate terms. It lists business cards, stationery, brochures, signs, rate lists and other promotional items as examples of that communication. An outbound call opener is a means of communicating. So is a text. So is a dropped voicemail.

The same section defines offering or negotiating terms to include communicating directly or indirectly with a borrower or prospective borrower for the purpose of reaching a mutual understanding about prospective residential mortgage loan terms. Directly or indirectly. A templated message sent on a rep’s behalf is indirect communication.

And the penalty follows the representation, not just the closing. Regulation H requires each state supervisory authority to have the legal authority to impose civil money penalties for individuals acting as loan originators, or representing themselves to the public as loan originators, in the state without a valid license or registration. Read that second clause twice. The conversation is the exposure.

So what do you actually build? Three things, and they are not exotic.

  • A license footprint field on every rep record, holding the list of states that person currently holds a license in, with an expiry date.
  • A property state field on every lead, populated at capture rather than guessed from the phone number.
  • A routing rule that will not connect a rep to a lead whose property state is missing from that rep’s footprint, plus a named fallback for the leads nobody on shift can take.

Kixie’s rule-based routing is included on all three plans, alongside ring groups and call queues, so the rule lives in the calling platform rather than in a separate tool somebody forgets to update. The failure mode is boring and common: the license list lives in a spreadsheet on a compliance manager’s desktop and the dialer has never read it. Renewals lapse, the spreadsheet updates, and the routing rule does not.

Put the footprint where the router can see it. Then test it. Send a lead from a state nobody on shift is licensed in and watch what the system does. Does it hold the lead, route it to a named fallback, or hand it to the next idle rep and hope? The answer to that one question tells you whether the rule is real or decorative.

Dialing modes a mortgage broker should actually test

Vendor category language is muddy on purpose. What actually separates these modes? Not the marketing label. Define each one by what the rep does on the floor and the fog clears.

  • Click to call. The rep reads the record, decides, and clicks. One call at a time, full context, slowest throughput.
  • Preview. The system queues the next record and shows it before dialing. The rep can skip. This is the mode that fits a licensed business, because the state is on screen before the line opens.
  • Power and progressive. The system advances through a loaded list automatically and connects the rep when someone answers. Throughput goes up. The decision point disappears.
  • Predictive. The system dials ahead of rep availability based on a pacing model. Highest throughput, and the mode that creates abandoned-call risk.

Kixie’s PowerDialer runs one line on the Single-Line PowerDialer plan and up to four lines on the Multi-Line PowerDialer plan, with calling session management on both. AI Human Voice Detection is a premium add-on available on the Multi-Line plan.

Now connect the line count to the license footprint, because this is where mortgage differs from a generic sales floor. Your dialable universe is not the list. It is the list filtered to the states the rep on the headset can legally originate in. A rep licensed in two states working a national lead file has a much shorter queue than the file size suggests. Buy line capacity against the filtered number, not the raw one.

That is also the honest argument for preview mode in a brokerage. It looks slower on a throughput chart. It puts the property state in front of a human before anything connects.

The general case for power dialing in a lending team stands on its own. The licensing filter sits on top of it, and it is the part that decides how much of the throughput you can actually use.

What the first mortgage message has to carry

Templates are where a licensed business scales its exposure without noticing. One text template, six reps, nine states, thousands of sends. The template does not know who is licensed where. Nobody checks it after launch week either.

Two different identification rules apply, and they do not cover the same people. Which one binds you? It depends on whether your originators are state licensed or federally registered, and most buying guides collapse the two. Keep them separate.

Regulation G governs mortgage loan originators who are registered rather than state licensed, meaning employees of covered financial institutions. For them, a registered mortgage loan originator must provide the unique identifier to a consumer upon request, before acting as a mortgage loan originator, and through the originator’s initial written communication with a consumer, if any, whether on paper or electronically. An outbound text is an initial written communication delivered electronically. If that is your structure, the identifier belongs in the template.

Regulation Z sets a narrower and more specific rule that applies to the loan documents themselves. A loan originator organization must include its name and NMLSR ID, plus the name and NMLSR ID of the individual originator with primary responsibility, on four documents: the credit application, the loan estimate and closing disclosure Regulation Z requires, the note or loan contract, and the security instrument. That list is closed. A text message is not on it, and nobody should claim otherwise.

So what is the operating rule? Decide the identification standard once, per state and per structure, and build it into the template set rather than into rep discretion. Then make the dialer pick the template by the lead’s property state, the same field the router reads.

Kixie includes business calling and texting on every plan, with SMS templates, Team SMS and Voicemail Drop. Voicemail Drop matters more than it looks here, because a drop is a recording a rep made, not a text field that merges. A state-specific drop is a separate recording. Plan for that in the setup, not in week three.

The failure mode is one national template naming a license nobody holds in the state it just landed in.

Local presence caller ID is not a mortgage license

Pickup rates drop when the number looks foreign to the person holding the phone. Local presence answers that by displaying a number matching the called party’s area code. It works. It is a real feature with a rule attached. It also gets oversold to this market, so draw the boundary clearly. Three rules bound it.

First, the FCC rule. Anyone who engages in telemarketing must transmit caller identification information. The only substitution the rule permits is the name of the seller on behalf of which the call is placed and the seller’s customer service telephone number, and the regulation adds a condition people skip: the telephone number so provided must permit any individual to make a do-not-call request during regular business hours. Telemarketers are also prohibited from blocking caller ID transmission.

That gives you a concrete demo question. When a borrower calls back the number your dialer displayed, during business hours, what happens? If it rings into nothing, or into a queue with no path to a do-not-call request, the displayed number is not doing the job the rule describes.

Second, the spoofing rule, stated accurately. No person or entity may, with the intent to defraud, cause harm, or wrongfully obtain anything of value, knowingly cause any caller identification service to transmit or display misleading or inaccurate caller identification information in connection with any voice service or text messaging service. The prohibition is bounded by intent. Displaying a local number is not itself the violation. Using a display to mislead someone about who is calling and why is the thing the rule reaches.

Third, the point that belongs to this market specifically. An area code is not a license. A Phoenix number on the screen does not make the originator licensed in Arizona, and it does not change which state’s rules the conversation falls under. If anything, a local display raises the odds that the person on the other end assumes a local, licensed relationship that may not exist.

Use local presence for what it is: a connection-rate tool with a callback obligation attached. Keep the do-not-call work separate and deliberate, and remember that the national registry is not the whole list, because several states run their own lists. Kixie’s automated DNC registry compliance is a compliance add-on rather than a standard plan inclusion, so price it in rather than assuming it.

What a mortgage dialer has to log for a state examiner

Here is the part buyers almost never put in the scorecard. Who else gets to read your call logs? A state examiner, if it comes to that. Your call data is examinable.

Pale lilac and pink scene in frosted glass: six blank rectangular glass plates of slightly different sizes fanned and overlapping in a low flat stack, one plate lifted and floating level above the others, with a small glass hemisphere resting apart on the right.

Regulation H requires a state supervisory authority to have the legal authority and mechanisms to examine any books, papers, records, or other data of any loan originator operating in the state, and to summon any loan originator or any person having possession, custody or care of the reports and records relating to that originator, to appear and produce those records and give testimony under oath. It also requires the supervisory authority to make a loan originator ensure that all residential mortgage loans closing as a result of its origination activity are included in reports of condition submitted to the NMLSR.

Your dialer holds a large share of those records. Call logs. Timestamps. Dispositions. Recordings. The number that was displayed. The rep who spoke. Could you produce all of that for a call from nine months ago, today, without asking a vendor for an export?

So write the log requirement into the evaluation and make the vendor show it on a live record, not a slide. At minimum you want to be able to answer six questions about any call from last quarter:

  • Which rep dialed it, by name, and which license footprint that rep held on that date.
  • Which number was displayed to the borrower.
  • Which state the property was in, carried on the record rather than inferred.
  • What the disposition was, in a controlled list rather than free text.
  • Whether a recording exists, where it is stored, and how long it is retained.
  • Whether all of it is still readable after that rep leaves the company.

That last one is the quiet killer. A deal, a recording and a call history that only exist inside a departed rep’s seat are not a record. They are a gap with a name on it. Ask the vendor what happens to a seat’s history on deactivation before you ask about pricing.

Kixie includes call recording with included cloud storage and bi-directional CRM integration on all three plans, and its webhook events cover start call, answered call, end call, disposition, call outcome, voicemail and SMS, so the record can be pushed into the system of record as it happens rather than reconciled later. The script and record rules on a refinance call go further into what the conversation itself has to contain.

How to run the mortgage dialer demo

Vendors will happily show you dial speed. Dial speed is not your constraint. What is? Whether the routing rule can read a license field. Drive the demo at that question and you will learn more in forty minutes than a two-week trial of the standard script will teach you.

Bring a test file with leads in three states and a rep record licensed in only one of them. Then ask for these, in order:

  • Show me where a rep’s licensed states are stored, and show me the routing rule reading that field.
  • Route a lead from a state this rep is not licensed in. What does the system do, and what does the rep see?
  • Show me the same rule applied to an outbound list load, not just an inbound route.
  • Pick the SMS template by the lead’s property state. Show me the merge actually changing.
  • Show me a voicemail drop and tell me what it takes to run a different one per state.
  • Show me the caller ID the borrower sees, then call that number back from an outside line while I watch.
  • Pull up a call from last month and show me every field in the six-question list above.
  • Deactivate a test seat and show me what survives.

Any vendor who can do eight of eight is worth a pilot. Most will stall on three, four and eight. Is that disqualifying? Not automatically. It tells you where the configuration work lands and who pays for it, and that is cheaper to learn in the demo than in month five.

A buying checklist for the best dialer for mortgage brokers

Score each item yes or no. No partial credit. Partial credit is how teams talk themselves into the wrong tool.

  • The platform stores a per-rep licensed-state list with expiry dates, and routing reads it.
  • Property state is a first-class field on the lead, captured rather than inferred from an area code.
  • Preview mode exists and shows the property state before the line opens.
  • Line capacity is sized against the license-filtered list, not the raw file.
  • Message templates can be selected by property state, and voicemail drops can vary the same way.
  • The displayed callback number reaches a human who can take a do-not-call request during business hours.
  • Do-not-call suppression covers state lists, not only the national registry.
  • Every call carries rep, displayed number, property state, disposition, timestamp and recording pointer.
  • A departed rep’s call history, recordings and deal context stay readable.
  • The CRM sync is bi-directional, so the record of the call is in the system of record automatically.
  • Your counsel has reviewed the campaign configuration before any automated outbound calling starts.

That last line is not a disclaimer. Consent, calling hours, suppression, recording notice, prerecorded voice and pacing all turn on facts about your campaign that a buying guide cannot see.

Pick the tool that passes the first ten and get the eleventh done before go-live. Then track one number for the first ninety days: how many leads were routed to a rep who could legally take them on the first attempt. If that number is not close to one hundred percent, the routing rule is wrong, not the reps.

Mortgage dialer questions brokers ask

Does a dialer need to know which states a loan officer is licensed in?

Functionally, yes, if you want the routing rule to hold. Regulation H ties the licensing requirement to any dwelling or residential real estate in the state, so the lead’s property state is what determines who may take it. Regulation Z then makes the loan originator organization responsible for ensuring the individual is licensed or registered before acting as a loan originator. The routing rule is where a firm either enforces that or leaves it to chance.

Can local presence caller ID get a mortgage broker in trouble?

The FCC’s prohibition on misleading caller ID is bounded by intent to defraud, cause harm, or wrongfully obtain anything of value, so displaying a local number is not itself a violation. The more practical rule is the telemarketing one: when you substitute a number, it must be the seller’s customer service number and it must permit any individual to make a do-not-call request during regular business hours. Test the callback path before you turn the feature on.

Does a text message count as a loan originator communication?

Regulation H counts representing to the public through advertising or other means of communicating or providing information that you can take an application or offer terms, and it counts communicating directly or indirectly to reach a mutual understanding about prospective loan terms. A template sent on a rep’s behalf is a means of communicating. Separately, Regulation G requires a registered mortgage loan originator to provide the unique identifier through the originator’s initial written communication with a consumer, whether on paper or electronically.

How many dialer lines should a mortgage brokerage run?

Size it against the license-filtered list rather than the file you bought. A rep licensed in two states working a national file has a far shorter dialable queue than the row count suggests, and extra lines against a short queue mostly produce repeat attempts. Kixie’s PowerDialer runs one line on the Single-Line plan and up to four lines on the Multi-Line plan.

What call records should a mortgage brokerage keep?

Keep enough that any call can be reconstructed by someone who was not there. Regulation H requires state supervisory authorities to be able to examine any books, papers, records or other data of any loan originator operating in the state, and to summon anyone with custody of those records. Rep identity, displayed number, property state, disposition, timestamp and recording location are the practical minimum, and all of it has to survive a rep leaving.

Nothing here is legal advice. The rules cited are current as of the review date below, and a qualified adviser should review your specific campaign configuration, consent records and state obligations before you start automated outbound calling.

Sources

  • Office of the Federal Register, Electronic Code of Federal Regulations, 12 CFR part 1008, S.A.F.E. Mortgage Licensing Act, State Compliance and Bureau Registration System (Regulation H). Section 1008.103(a) requires a state to prohibit an individual from engaging in the business of a loan originator with respect to any dwelling or residential real estate in the state unless the individual first registers through the NMLSR and obtains a unique identifier, and obtains and maintains a valid state loan originator license. Section 1008.103(b)(2) includes representing to the public, through advertising or other means of communicating or providing information, that the individual can or will perform those activities. Section 1008.103(c)(2)(i)(B) defines offering or negotiating terms to include communicating directly or indirectly with a borrower or prospective borrower to reach a mutual understanding about prospective terms. Section 1008.111(b)(1) and (b)(2) give the state supervisory authority power to examine any books, papers, records or other data of any loan originator operating in the state and to summon any person with custody of those records. Section 1008.111(b)(6) requires authority to impose civil money penalties on individuals acting as, or representing themselves to the public as, loan originators in the state without a valid license or registration. Section 1008.111(f) requires loans closed through origination activity to be included in reports of condition submitted to the NMLSR. https://www.ecfr.gov/current/title-12/chapter-X/part-1008. Reviewed November 15, 2026.
  • Office of the Federal Register, Electronic Code of Federal Regulations, 12 CFR 1007.105, Use of unique identifier (Regulation G). Paragraph (a) requires a covered financial institution to make its registered mortgage loan originators’ unique identifiers available to consumers. Paragraph (b) requires a registered mortgage loan originator to provide the unique identifier to a consumer upon request, before acting as a mortgage loan originator, and through the originator’s initial written communication with a consumer, if any, whether on paper or electronically. https://www.ecfr.gov/current/title-12/chapter-X/part-1007/section-1007.105. Reviewed November 15, 2026.
  • Office of the Federal Register, Electronic Code of Federal Regulations, 12 CFR 1026.36, Prohibited acts or practices and certain requirements for credit secured by a dwelling (Regulation Z). Paragraph (f)(2) requires a loan originator organization to ensure that each individual loan originator who works for it is licensed or registered to the extent required under the SAFE Act licensing rules and state implementing law before the individual acts as a loan originator in a consumer credit transaction secured by a dwelling. Paragraph (g) requires the organization’s name and NMLSR ID and the individual originator’s name and NMLSR ID on the credit application, the loan estimate and closing disclosure Regulation Z requires, the note or loan contract, and the security instrument. https://www.ecfr.gov/current/title-12/chapter-X/part-1026/subpart-E/section-1026.36. Reviewed November 15, 2026.
  • Office of the Federal Register, Electronic Code of Federal Regulations, 47 CFR part 64 subpart P, caller identification requirements. Section 64.1601(e) requires any person or entity that engages in telemarketing to transmit caller identification information, permits substitution only of the name of the seller on behalf of which the call is placed and the seller’s customer service telephone number, requires that number to permit any individual to make a do-not-call request during regular business hours, and prohibits telemarketers from blocking caller identification transmission. Section 64.1604(a) prohibits knowingly causing a caller identification service to transmit or display misleading or inaccurate caller identification information in connection with any voice service or text messaging service with the intent to defraud, cause harm, or wrongfully obtain anything of value. https://www.ecfr.gov/current/title-47/chapter-I/subchapter-B/part-64/subpart-P. Reviewed November 15, 2026.

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.

Best Texting Software for Insurance Agents Keeps the Book

TL;DR: The best texting software for insurance agents is decided by three things a demo almost never shows. First, consent does not travel. CTIA’s Messaging Principles say a consumer opt-in should not be transferable or assignable and should apply only to the campaigns and the specific message sender it was obtained for, so an acquired book of business arrives with no permission to text it, and rented or purchased opt-in lists are out. Second, a health, Medicare or group benefits book can pull the vendor inside the HIPAA Privacy Rule. If your agency creates, receives, maintains or transmits protected health information on behalf of a health plan, and the Privacy Rule counts a health insurance issuer, an HMO, Part D and a Medicare supplemental policy issuer as health plans, then your texting vendor is a subcontractor and the assurances have to be documented in a written contract. Most property and casualty books never touch this. Third, the number and the thread are agency assets, not producer assets, so a shared number with a shared inbox beats a producer’s personal cell every time a producer leaves. After that, test the boring parts. Does STOP on one agency number suppress every other number you own? Does the platform read ‘please take me off your list’ or only the literal keyword? Does a voice opt-out a service rep logged on Tuesday stop Thursday’s renewal text? The FCC’s do-not-call rules give you a reasonable time to honor an internal request and cap it at 10 business days, which is a propagation requirement, not a policy sentence. Then check that the thread writes back to the agency management system rather than a CRM nobody opens, because insurance texting is a service queue before it is a campaign tool.

A producer texts a quote from a personal cell. Nine months later the client replies about a claim. The producer left for a competitor in March and took the phone with them. The agency has the policy, the carrier has the claim, and nobody has the conversation.

That is the failure the best texting software for insurance agents actually has to prevent, and it is not on any feature grid. Most comparison pages rank two-way messaging, templates, and automation, and those things do matter, but they ship in every product on the market at this point, which means a side-by-side on them produces a tie and tells you nothing about which tool survives contact with an agency.

Three things do decide it: whose permission you are texting on, whether the book you write drags a federal privacy contract into the purchase, and who keeps the number and the history when a person leaves. Get those wrong and the feature grid is decoration. Nothing else on the page saves it.

What the best texting software for insurance agents has to settle first

Start with a boring inventory. It takes ten minutes. Write down every population the agency texts: new leads from a purchased source, quoted prospects who have not bound, current policyholders during service, policyholders at renewal, claimants mid-claim, and the Medicare-eligible segment if you carry one. Six populations. They do not share a consent posture, a record obligation, or a reply volume.

Six populations, six different answers to the only question that matters, which is whether you are allowed to send this message to this person today. Now ask the three questions that separate the shortlist.

  • Whose campaign is this. The messaging program is registered to a brand. If that brand is not your agency, you are renting permission you cannot move.
  • Does the book reach protected health information. A health, Medicare or group benefits book can turn the vendor into a subcontractor under the Privacy Rule. A property and casualty book usually does not.
  • Who owns the number and the thread. If the answer is “the producer,” you do not have a texting system. You have a staffing risk.

Everything else on the evaluation is real, and conversation search, assignment, templates, reporting, number management and export all deserve a score, but they are second, because a product that wins on all six and fails any of the three above is a product you will be replacing inside a year.

Insurance texting consent does not transfer with a book of business

Agencies grow by acquisition. A principal buys a retiring agent’s book, imports the contacts, and starts a renewal text campaign the following Monday. That campaign is the single most common way an agency breaks its own messaging program.

Pale violet scene rendered in frosted translucent glass. A wide squat canister is filled with a stack of flat plates, an upright glass pane stands tilted beside it, a single small sphere is stopped against the foot of the pane, and a tall narrow canister on the other side is completely empty.

CTIA’s Messaging Principles and Best Practices is explicit about it. A consumer opt-in “should not be transferable or assignable,” and it “should apply only to the campaign(s) and specific Message Sender for which it was intended or obtained.” The same document says message senders should not use opt-in lists that have been rented, sold, or shared, and should create and vet their own.

Read that against how insurance actually buys growth. The retiring agent’s clients consented to the retiring agent, not to you, and the shared lead vendor’s list is by definition a shared list, so neither one arrives with permission attached no matter what the purchase agreement says about the transfer of customer records. The records transfer. The consent does not.

So what does that make a software requirement rather than a compliance memo? Three concrete things you can check on a trial.

  • Consent has to be a field on the contact, with a source and a date, not an assumption baked into a list upload.
  • An import has to be able to land in a non-messaging state so an acquired book cannot be campaigned by accident on day one.
  • The platform has to let you run a permission-gathering sequence on a channel the acquired client already agreed to, usually mail, email, or a call, before the first text.

Test it rather than ask about it. Import 50 fake records with the consent field empty and try to campaign them, because if the system lets you blast immediately with nothing populated, it will let a new hire who has never read this do exactly the same thing in month four and the first you hear about it is a carrier complaint. Purchased leads carry their own separate problem, and the rules around purchased insurance leads are worth reading before you pay for another batch.

The health insurance question that changes which texting software you can use

This is the section most insurance texting comparisons skip entirely, and it is the one that can disqualify a vendor outright before price, features or integrations ever come up.

The HIPAA Privacy Rule defines a business associate as a person who, on behalf of a covered entity, creates, receives, maintains, or transmits protected health information for a regulated function, including benefit management. It defines health plan to include a health insurance issuer, an HMO, Part A or Part B of Medicare, the Part D prescription drug program, an issuer of a Medicare supplemental policy, and an issuer of a long-term care policy other than nursing home fixed indemnity.

Then it closes the loop. A business associate may let a subcontractor create, receive, maintain, or transmit protected health information on its behalf only if it obtains satisfactory assurances that the subcontractor will safeguard the information, and those assurances have to be documented through a written contract.

Translate that into the purchase. If your agency is handling protected health information on behalf of a health plan, the texting platform carrying those messages is the subcontractor. The written agreement is not a line item on a security questionnaire. It is the condition of using the product for that book at all. No contract, no texting.

Draw the boundary honestly, because it is narrower than vendors selling fear would like it to be. A personal lines agency writing auto and homeowners is generally not touching protected health information and generally does not need any of this, while a health and Medicare shop very well might, and a mixed agency has to decide which book runs on which tool. That decision is a lot cheaper to make before the contract than after.

Ask one question on the vendor call. Will you sign a business associate agreement, and on which plan tiers? A vendor that has never been asked will not have an answer ready, and the shape of the non-answer tells you how much of their book looks like yours. The Medicare marketing rules that bind agent outreach sit next to this and are worth understanding at the same time.

Who keeps the number when a producer leaves the agency

Walk the queue. A producer collects a mobile number at the kitchen table. The producer texts from a personal cell, because it is faster and nobody stopped them. The client replies there for two years. The producer resigns. The agency sends a retention letter to an address while the client is texting a number the agency does not control, and gets no answer, and calls it churn.

That is not a loyalty problem. It is a records problem wearing a loyalty costume. Fix the custody and the retention number moves on its own.

The fix is structural, not motivational, and it comes down to three conditions holding at once: the number belongs to the agency, the thread lands somewhere more than one person can open it, and reassigning the owner when a producer leaves moves the queue without touching the history.

Kixie includes business texting tied to the same number your team calls from on all three plans, along with Team SMS shared inboxes and SMS templates, so the conversation is an agency record from the first message rather than something you try to recover later. Whatever you buy, check that two conditions hold: a second user can read the full thread without the first user’s device, and reassigning an owner does not orphan the history.

There is a quieter version of the same failure. Some agencies give each producer their own dedicated number inside the platform, which is fine, until the producer leaves and the number is reassigned to a new hire who inherits two years of context they have never read. Decide in advance whether a departing producer’s number gets reassigned, parked, or forwarded, and write it down. The platform will not decide it for you.

Insurance texting software has to write back to the agency management system

Here is where insurance diverges from every other vertical buying texting software. A car dealership runs on a CRM. A home services shop runs on a field service platform. An insurance agency runs on an agency management system, and most sales texting tools were not built against one.

That is not automatically a problem, but it is a scoping question you have to ask out loud on the first call, because the difference between a native connector and a custom build changes the implementation budget and the person who owns it by an order of magnitude.

So ask it directly. Is my agency management system a native connector in your product, or is it a build? If it is a build, what is the mechanism, who does the work, and what breaks when it fails? A vendor with a documented pattern for non-native systems, usually webhooks plus an automation tool or a direct API, is in a much better position than one that says “we can integrate with anything” and leaves.

Then decide what actually has to sync. The honest answer is usually less than the demo suggests.

  • The message body and direction on the client record, so the next person to open the account sees the conversation.
  • The consent state, so an opt-out recorded in the texting tool reaches the system that drives renewal campaigns.
  • The timestamp, because a claim dispute is a timeline argument before it is anything else.

Attachments, read receipts and delivery codes are nice, and none of them are what a records request asks for, so scope the sync to what you would actually have to produce and then test it against a real account rather than a sandbox contact named Test Test.

Opt out handling is where insurance texting software quietly fails

Every vendor will tell you they handle opt-outs, and almost every one of them means that they catch the literal keyword on the one number that received it, which is a fraction of the obligation and the fraction that is easiest to build.

Pale violet diagram rendered in frosted translucent glass. Three horizontal glass tubes of different diameters enter from the left. The lowest and widest tube is sealed with a solid plug and holds three spheres stopped behind it, while spheres stream freely out of the two open tubes above and collect in a shallow glass basin on the right.

CTIA’s guidance is broader than the keyword. Message senders should support multiple opt-out mechanisms including phone call, email, or text. Standardized STOP wording should be used in the instructions, but requests in normal language should also be read and acted on, and the document names stop, end, unsubscribe, cancel, quit, and “please opt me out.” It adds that the validity of an opt-out should not be affected by small variances such as capitalization, punctuation, or letter case. One final confirmation message per campaign, then nothing further.

The FCC’s do-not-call rules put a clock on the internal list separately. A request not to receive calls from that entity has to be honored within a reasonable time, and that period may not exceed 10 business days from receipt.

None of that is hard to summarize and all of it is easy to get wrong, so turn it into four things you can test in an afternoon instead of four things you can promise in a policy document.

  • Text STOP to one agency number. Try to send from a different agency number an hour later. Did it go out?
  • Reply with “please take me off your list” instead of the keyword. Did the platform catch it, or did it answer with a renewal reminder?
  • Have a service rep log a verbal opt-out on a call. Does the texting side know?
  • Send the confirmation, then trigger an automated campaign that contact is enrolled in. Does the suppression hold?

Agencies fail the second and third tests constantly, and both failures look identical from the outside. A client asked you to stop. A text went out anyway. Now the brand is suspect on every channel, including the one that would have renewed the policy. The consent-type breakdown for insurance text messages covers which message types each permission level actually supports.

Insurance texting is a service queue before it is a campaign tool

Most texting software is sold on outbound. Insurance uses it inbound, and the best texting software for insurance agents is judged on the inbound side first. A client texts a photo of a windshield at 7 in the morning, asks whether a teenage driver is covered, or wants to know if the payment went through before a renewal lapses.

That changes the specification completely. Send throughput matters less than reply time and automation matters less than routing, because the product question for a service-heavy book is not how many messages can I send, it is who sees this inbound message in four minutes and what happens when nobody does.

So measure the queue, not the blast. Three numbers are enough to run on.

  • Median first reply time during business hours, and the outliers separately. The median hides the renewal that waited two days.
  • Unassigned conversations older than one hour. If this number is not zero at 5 p.m., you have a staffing gap, not a software gap.
  • Threads with no next step. A conversation that ended without a task, a quote, or a note is the texting version of a closed-lost with two lazy attempts behind it.

Staff it before you scale it, because an agency that turns on inbound texting without naming the person who watches the queue has quietly promised its entire book a response channel that nobody owns, and the client who texted about a windshield at seven in the morning does not know that.

How to test texting software for insurance agents in one week

You do not need a 90-day evaluation of the best texting software for insurance agents. You need five days and a willingness to break things.

  1. Day one, registration. Confirm the messaging brand registers to your agency’s own legal entity, not a carrier’s and not the vendor’s. Ask what happens to the number if you cancel.
  2. Day two, consent. Import 50 test records with no consent field set and try to campaign them. Then import an “acquired book” and confirm you can land it in a non-messaging state.
  3. Day three, opt-outs. Run the four tests above. Record which ones the platform passes without configuration and which need a rule you have to remember to build.
  4. Day four, the system of record. Push a real conversation to your agency management system. Then delete it on the texting side and see whether the record survives.
  5. Day five, the departure drill. Deactivate a test user mid-conversation. Can a manager read and continue the thread? Does the number stay with the agency?

Score demonstrated behavior only. A roadmap commitment is worth exactly nothing in an evaluation, and insurance buyers hear more of them than most, because the vertical is attractive and under-served and almost every vendor in it will tell you an agency management integration is coming next quarter.

If a product passes all five days, the feature grid will take care of itself, and if it fails day one or day five you can stop the evaluation there and keep the week.

Questions insurance agents ask about texting software

Can an agency text a book of business it just acquired?

Not on the strength of the acquisition alone. CTIA’s guidance says an opt-in should not be transferable or assignable and should apply only to the campaigns and the specific message sender it was obtained for. The practical path is to re-permission the acquired book on a channel those clients already agreed to, then text the ones who say yes.

Does an insurance agency need a business associate agreement with its texting vendor?

It depends on the book. If the agency creates, receives, maintains, or transmits protected health information on behalf of a health plan, the Privacy Rule treats the texting vendor as a subcontractor and requires the assurances to be documented in a written contract. A personal lines agency writing auto and homeowners generally is not in that position. Ask your own counsel before you decide which side of the line your agency sits on.

Should producers text clients from personal phones?

No, and the reason is custody rather than etiquette. A thread on a personal device is not an agency record, cannot be reassigned, and leaves with the producer. Use an agency-owned number with a shared inbox so a second person can read and continue the conversation.

What happens to text history when a producer leaves?

On an agency-owned number with a shared inbox, nothing. The queue is reassigned and the history stays. On a personal cell, the history is gone, and the client keeps texting a number that no longer reaches you. Decide in advance whether a departing producer’s number is reassigned, parked, or forwarded.

How fast does an opt-out have to be honored?

The FCC’s rules require an internal do-not-call request to be honored within a reasonable time, not to exceed 10 business days from receipt. In practice the number to test is propagation, not policy. If STOP on one agency number does not suppress the others, 10 days is irrelevant because the suppression never arrived.

What is the one test most agencies skip?

The voice-to-text handoff. Log a verbal opt-out on a service call, then try to send that same contact an automated renewal text and watch what happens. Almost nobody tests it. It is where the complaint comes from.

Sources

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.

Gemini SMS for Sales Teams and What It Cannot Do

TL;DR: Gemini SMS is not one feature, it is two, and Google’s own documentation rules both of them out as the sender for business texting. The Gemini mobile app on Android can send a text. It sends through the Messages app using the phone’s default messaging app, so the carrier connection and the originating number belong to the handset in the rep’s pocket. It also needs an active mobile phone plan, the SMS Android permission granted to the Google app, and contacts saved to the user’s Google Account, and by default it routes through the last app used to message that contact. A prospect list is not an address book. The other surface, Gemini inside Google Messages, requires a personal Google Account you manage on your own, and Google states you cannot use it with a Google Workspace account or one managed by Family Link. That is the account your sales team runs on. It also needs Android, a supported language, the current Google Messages app, RCS chats turned on, and an age over eighteen, and Google says the rollout is gradual across more than 165 countries. Neither surface handles the reply. Google publishes the unsupported actions itself, and the list is short. No reading or summarizing messages from your messaging app, no group chats, no reading images, no audio or video, and no reading or responding to notifications. Outbound is the easy half. Google also says messages created with the Gemini app are not end-to-end encrypted, tells you not to enter confidential data while Keep Activity is on, and keeps a reviewed subset of chats for up to three years even after you delete your activity. CTIA draws the last line. A consumer using automation to compose and send their own messages is typical consumer operation, and automation used by a non-consumer to facilitate messaging is not, so the same prompt changes classification based on who is sending. Use Gemini to draft, shorten, classify and summarize. Send from a registered business number, log the thread on the contact record, and handle the opt-out there.

What Gemini SMS Means on Android

Most people searching for Gemini SMS are not asking a product question. They are asking whether the assistant on their phone can type the text for them and press send. The answer is yes, under conditions, and the conditions are the whole story.

Gemini SMS is a search phrase rather than a product name. Google ships two separate things behind it, and they do not behave the same way.

  • The Gemini mobile app on Android, which can compose a message and send it through a connected Messages app.
  • Gemini inside Google Messages, which sits in the app as a chat partner and helps you draft, brainstorm, or plan.

One of them sends. The other one talks. Google’s support documentation is explicit that the Messages app is not supported in Gemini in Google Messages or in the Gemini web app, so the sending path and the in-thread path are genuinely different products with different requirements.

That distinction is not pedantic. A rep who reads one article about Gemini drafting a reply and assumes the same thing will fire off a prospecting text from a work laptop is going to be wrong twice.

Can Gemini Send SMS Messages for You

Yes, from the Gemini mobile app on an Android phone or tablet, and Google lists exactly what has to be true first.

  • An Android phone or tablet.
  • An active mobile phone plan.
  • The Gemini app, including as your mobile assistant.
  • The setting that saves contact info from your signed-in devices, turned on.
  • The SMS Android app permission granted to the Google app.

Read the third and fifth items together and the architecture becomes obvious. Gemini SMS is not a messaging service. It is a layer that drives your handset. Google says the Messages app in Gemini uses your phone’s default messaging app, which means the carrier connection, the originating number, and the delivery all belong to the phone in your pocket.

So whose number sends the text? Yours. The personal one. The number a prospect will call back, save, or report.

There is a routing detail worth knowing before anyone tries this on a customer. Google says that by default the Gemini mobile app uses the last app it used to message the same contact. The route is inherited from history, not chosen by you. If a rep has ever reached that contact through a different messaging app, the next send can follow the old path.

Contact resolution is the harder constraint. Gemini sends to contacts saved to your Google Account, which is why the contact sync setting is in the requirement list at all. Google adds that allowing the Contacts permission lets it reach device contacts that are not synced. Either way, the universe of recipients is your address book.

A sales list is not an address book. It lives in the CRM, it changes weekly, it carries ownership and consent state, and nobody wants four hundred prospects synced into a rep’s personal Google contacts. That is the first hard stop, and it has nothing to do with how good the model is.

Why Gemini SMS Is Blocked on a Google Workspace Account

The other surface is the one people usually mean when they say Gemini SMS. It lives inside Google Messages and helps draft and brainstorm in the thread. It also has the single most decisive requirement in this entire article.

Google’s requirements list says you need a personal Google Account that you manage on your own, and then states plainly that you cannot use Gemini in Google Messages with a Google Account managed by Family Link or with a Google Workspace account.

Your sales team runs on Workspace. That is what the company email is. So the Gemini surface that sits inside the conversation is the one your work identity is specifically not allowed to open.

The rest of that requirement list adds friction rather than blocking:

  • An Android phone, so no iPhone and no desktop.
  • The phone language set to a supported language, from a list of roughly forty.
  • The latest version of Google Messages.
  • RCS chats turned on.
  • Being 18 or over.

Google also says the feature is being released gradually and, as of November 2026, lists it as available in over 165 countries. Gradual release is the quiet part. It means two reps on the same team, same phone model, same month, can see different behavior, and neither one is doing anything wrong.

The RCS requirement deserves one sentence of its own, because it moves the feature onto a transport that can simply be unavailable. Google documents four RCS statuses, including trouble connecting and not supported, and says that if RCS chats are disabled by your carrier you have to contact the carrier. A feature gated on a carrier setting is not a feature you can roll out to a floor.

What Gemini SMS Cannot Do in Messages

Google publishes an explicit list of unsupported actions, and it is more useful than any review. These are the things Gemini cannot do with the Messages app:

Pale violet scene with a frosted glass funnel pouring three small glass spheres out through an open glass aperture ring, beside a matching aperture ring sealed flush with a frosted glass disc holding two spheres stopped against its face.
  • Read or summarize your messages directly from your messaging app.
  • Create or message group chats.
  • Read images or memes in your messages.
  • Add or play audio or videos in your messages.
  • Read or respond to messages in your device notifications.

Look at the first and last items. There is no inbound. Gemini can push a message out through your handset and then it is blind to what comes back, unless you connect a separate device assistance app that reads notifications, which Google scopes as a different capability entirely.

That matters more than the send does. Outbound is the easy half of texting. The work is the reply: reading the response, classifying it, routing it to the right rep, honoring an opt-out, and logging the whole exchange against the deal. A Gemini SMS flow that writes the first message and then goes quiet has automated the fastest part of the job.

The group chat limit closes another common assumption. Anyone imagining a broadcast to a list is imagining a product that does not exist here.

One more line from the same page is worth keeping in front of a security review. Google says that even if you disconnect the Messages app, the Gemini mobile app can still access your device contacts when the Google app holds the Contacts permission. Disconnecting the integration does not revoke the contact access, and those two settings live in different places.

Where Gemini SMS Conversations Are Stored

Every Gemini SMS a rep asks for starts life as a prompt, and a prompt is a chat. Google’s privacy documentation describes what happens to chats, and it is written clearly enough that there is no room to guess.

Start with the sentence that should end most internal debates. Google says a subset of chats are reviewed by human reviewers to help improve its services, and that reviewed chats are retained for up to three years. It then gives direct instruction: if Keep Activity is on, do not enter data that is confidential or that you would not want a reviewer to see.

A prospect’s name, mobile number, employer, and the reason they are in your pipeline is confidential customer data. The instruction is not ambiguous.

The retention mechanics are worth holding onto:

  • Gemini Apps Activity defaults to an 18 month auto-delete, changeable to 3 months, 36 months, or indefinite.
  • Chats reviewed by human reviewers are not deleted when you delete your activity, and are kept up to three years.
  • Temporary chats and chats recorded while Keep Activity is off are retained for 72 hours, with the last 24 hours used as context, and are not used to train the models.
  • Google says chats are disconnected from your account before being sent to service providers.

The second item is the one that breaks a records policy. Your deletion does not reach the reviewed copy. If your team has a data retention commitment in a customer contract, a rep pasting account context into an assistant has created a copy with its own clock.

Connected Apps collect more than the prompt. Google lists Gemini Apps activity, device calling and messaging logs, contact info, and device info and settings as the data processed when you use Connected Apps to make calls or send messages. It also notes that up to 90 days of saved Google Assistant communications request history may be imported into Gemini Apps Activity for anyone who used Assistant for this before.

And the encryption line, stated twice in Google’s own documentation: messages that you create and send with the Gemini app are not end-to-end encrypted. For a personal text to a friend, fine. For a message carrying a customer’s financial or health context, that is a sentence a compliance officer will want to read before the feature ships.

Why Sales Teams Need More Than Gemini SMS

Set the product aside for a moment. The carriers already drew this boundary, and they drew it by behavior rather than by intent.

Pale violet scene with a slender frosted glass capillary tube packed solid with tiny glass beads running on a steep diagonal, beside a short wide-bellied glass vessel holding three widely spaced large glass spheres.

CTIA’s Messaging Principles and Best Practices describes the attributes of typical consumer operation. The numbers are the useful part:

  • Throughput of roughly 15 to 60 messages per minute.
  • Volume around 1,000 messages per day.
  • One telephone number assigned to or used by a single consumer.
  • Around 100 distinct recipients per message.
  • A roughly even ratio of outgoing to incoming messages per telephone number.
  • No substantially repetitive messages.

Now hold a prospecting block against that list. The balance attribute is the one that fails first. Consumer traffic is conversational, so an outgoing message usually produces a reply. A rep working a list sends many and hears back from few, and the ratio on that number stops looking like a person within a day.

CTIA then says the part that decides this article. A consumer using automation to help compose and send their own messages falls within typical consumer operation. Automation used in whole or in part by a non-consumer to facilitate messaging does not. Same assistant, same prompt, different classification, because classification follows who is sending and why.

So a rep using Gemini SMS to text a prospect from a personal handset is a business running automation on a consumer rail. That is precisely the traffic pattern carrier filtering exists to catch. Nobody gets a warning. Messages just stop arriving, and the rep keeps working a list that has quietly gone dark.

Business texting runs on the other rail: a registered business number, a sender identity the carrier can see, documented consent, and an opt-out path. Kixie’s side of that is business text messaging from the same number your team calls from, with the thread attached to the contact record. The registration layer is its own topic and is covered in the 10DLC compliance guide. The permission question, which tier of consent a given prospect text needs, is covered in the consent tier breakdown. Neither one is optional and neither one is something an assistant can supply.

How Sales Teams Can Use Gemini SMS Safely

None of this makes Gemini SMS useless for a sales team. It makes the send button the wrong place to put it.

The pattern that works is boring and it is the one that survives an audit. Gemini does the language work. Your system of record does the sending, the logging, and the opt-out.

  • Draft and shorten. A rep pastes the situation with the identifying details stripped and asks for three versions under 160 characters. Nothing customer-identifying leaves the CRM.
  • Classify the reply. An inbound text is routed into a workflow that asks Gemini to tag it as interested, not now, wrong person, or stop, and a human confirms anything that reads like a revocation.
  • Summarize the thread. After a call, the transcript and the texts become a short account note that the next rep can actually read.
  • Draft the sequence, not the send. The assistant writes the three-touch pattern; the platform schedules it against the number that is registered to send it.

Mechanically, Gemini connects to Kixie the way every AI connection does, as a custom build. Kixie has no native Gemini integration, so the connection runs over Kixie webhooks, Zapier actions, or the Kixie API. Kixie’s webhook events include Start Call, Answered Call, End Call, SMS, Disposition, Call Outcome, Voicemail, Scheduled Activity, and the Conversation Intelligence summary, which is the stream an assistant can read. On the other side, Zapier exposes a Send SMS action and the PowerList and Cadence actions, which is how a Gemini decision turns into a text sent from the business number rather than the rep’s handset. The Kixie API has no direct inbound auto-call endpoint, so an outside trigger is brokered through Zapier rather than firing natively.

If you want a worked example of this shape on a different model, the ChatGPT SMS bot walkthrough builds the same split end to end.

Check four things before you call it done. Does the message leave from a number the business owns? Is the thread on the contact record? Does a stop reply register as an opt-out without a human remembering? Can another rep pick up the conversation tomorrow? If any answer is no, the assistant is not the problem. The rail is.

Gemini SMS FAQs for Sales Teams

Does Gemini SMS work on iPhone?

No. Google’s requirements name an Android phone or tablet for sending through the Messages app, and an Android phone for Gemini in Google Messages. There is no documented iPhone path for either one.

Can Gemini send a text without confirmation?

It sends as soon as you ask it to, so your own prompt is the confirmation rather than a separate approval step. Google documents the send as something you ask for conversationally, with prompts like text a contact a message, and the on-screen flow handles connecting the Messages app the first time. Treat every outbound Gemini SMS as something a person should read before it goes, because the recipient match comes from your contacts and the routing follows the last app used for that contact.

Can Gemini read my text messages?

Not through the Messages app integration. Google lists reading or summarizing messages from your messaging app, and reading or responding to messages in notifications, as actions the Messages app in Gemini does not support. Notification reading is handled by a separate device assistance connection.

Can sales teams use Gemini for prospecting texts?

Not as the sender. Gemini in Google Messages excludes Google Workspace accounts outright, the mobile app sends from the rep’s personal number and personal contacts, and CTIA treats automation used by a business to facilitate messaging as outside typical consumer operation. Use Gemini to draft and classify, and send from a registered business number.

Why is Gemini texting not working?

Almost always a missing requirement rather than a fault. Work the list in order. Check that the Google app has the SMS permission, that contact info from your signed-in devices is saved, that the contact exists in your Google Account, and that you are on the mobile app rather than the web app, which does not support the Messages app at all. For Gemini in Google Messages, check that the account is personal rather than Workspace and that RCS chats show a connected status.

Is a message sent through Gemini encrypted?

Google states that messages you create and send with the Gemini app are not end-to-end encrypted. The underlying transport still applies, so an RCS chat and a plain SMS have different properties, but the Gemini documentation makes no end-to-end claim for messages it composes.

Sources

  • Google, Send messages with the Gemini mobile app on Android, for the requirement list, the use of the phone’s default messaging app, the contact and permission conditions, the default routing behavior, the unsupported-action list, and the statement that messages created with the Gemini app are not end-to-end encrypted. https://support.google.com/gemini/answer/15574940. Reviewed November 15, 2026.
  • Google, Use Gemini in Google Messages, for the personal Google Account requirement, the exclusion of Google Workspace and Family Link accounts, the RCS and age requirements, the supported language list, and the gradual release across more than 165 countries. https://support.google.com/gemini/answer/14599070. Reviewed November 15, 2026.
  • Google, Gemini Apps Privacy Hub, for human review and the three-year retention of reviewed chats, the instruction not to enter confidential data while Keep Activity is on, the default 18 month auto-delete, the 72 hour temporary chat retention, the Connected Apps data categories, and the 90 day Assistant history import. https://support.google.com/gemini/answer/13594961. Reviewed November 15, 2026.
  • Google, Turn on RCS chats in Google Messages, for the RCS chat statuses, carrier-disabled behavior, and how the compose bar indicates whether a message sends by data, SMS, or MMS. https://support.google.com/messages/answer/7189714. Reviewed November 15, 2026.
  • CTIA, Messaging Principles and Best Practices, the voluntary industry document dated May 2023, for the attributes of typical consumer operation covering throughput, volume, sender, recipients, balance and repetition, and for the statement that automation used by a non-consumer to facilitate messaging is not typical consumer operation. https://api.ctia.org/wp-content/uploads/2023/05/230523-CTIA-Messaging-Principles-and-Best-Practices-FINAL.pdf. Reviewed November 15, 2026.

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.

Can ChatGPT Make Phone Calls? Your Carrier Still Dials

TL;DR: Can ChatGPT make phone calls is really three questions, and only one of them ends with a prospect’s phone ringing. Talking to ChatGPT in the app is a voice session inside software, not a call on the phone network. OpenAI does support real telephony, but through the Live API over a SIP trunk, and its own telephony guide hands four separate jobs to someone else. You buy the phone number from a provider such as Twilio, you configure the trunk, your provider places the connection, and your provider finishes any transfer. The outbound request wants a full international phone number in transport.destination, SIP URI destinations are refused, and you supply the caller ID yourself in transport.trunk.caller_number. Outbound SIP calling runs through the Live API and not the Realtime API call-creation endpoint. The hard limits are published and not configurable, so ringing stops at 3 minutes, a connected call stops at 2 hours, and the sideband event stream only replays the preceding 3 seconds, so a late attach misses the ringing and answered events entirely. OpenAI also tells you to treat inbound SIP headers as untrusted caller metadata rather than authorization. None of that makes the model dial. It makes the model talk while a telephony platform does the calling, the number reputation, the recording, and the logging. Kixie states plainly that it ships no native ChatGPT integration, so teams wire the two together with Kixie webhooks, the Kixie API, or Zapier, Make and n8n, and push call events such as Answered Call, Disposition, Call Outcome, Voicemail and SMS into ChatGPT for summaries, follow-up drafts, and CRM notes, while any call or text fired from the ChatGPT side is brokered through Zapier because the Kixie API has no direct auto-call endpoint.

Search this question and you get a pile of posts from December 2024 about a consumer hotline. That is not a useless answer. It is just the wrong direction.

You dialing a number to talk to an assistant is not the same as an assistant dialing your prospect. One of those is a novelty. The other one touches your phone numbers, your consent records, your CRM, and your pickup rate.

So let’s separate what ChatGPT actually does from what the phone network actually requires.

Can ChatGPT Make Phone Calls Or Not

Three different things get called “ChatGPT making a phone call,” and they are not interchangeable.

The first is in-app voice. You speak, the model speaks back, and no phone number is involved anywhere in the chain. The second is a consumer calling feature, where a person picks up their own phone and reaches an assistant. The direction there is inbound, toward the assistant.

The third is the one that matters for outbound sales. A system places a call over the phone network to a number you chose, and the model handles the conversation when somebody answers.

That third one is real, and OpenAI documents it. Its telephony guide describes connecting incoming and outgoing phone calls to GPT-Live through a SIP trunk, with TLS for signaling and SRTP required for call audio. The guide splits into two flows: handle an inbound call, or place an outbound call.

So the short answer is yes, with a condition attached that changes everything downstream. The model does not own a phone line. You bring one.

ChatGPT Voice Chat Is Not a Phone Call

This distinction sounds pedantic until a rep tries to build on it.

Voice chat lives inside an application. The app captures audio, the model responds in audio, and the whole exchange stays in software. OpenAI’s own architecture guide frames the Realtime API around a browser session: your server mints an ephemeral client secret, the frontend opens a session, and the connection runs over WebRTC in the browser or a WebSocket on the server.

Nothing in that list is a phone call. There is no number, no carrier, no ring, no voicemail, no busy signal.

A phone call adds an entire layer the browser never had to think about. Something has to hold a phone number, reach the carrier network, negotiate audio, and survive the dozen boring ways a real call ends badly. That layer has a name in the docs and it is not the model.

Here is the practical test. If a rep can hear the AI talk but cannot point at the phone number it called from, there is no telephony in the setup yet.

What It Takes to Make ChatGPT Phone Calls Happen

Read OpenAI’s telephony guide closely and one phrase keeps repeating: your provider.

Four frosted glass shapes in a row on a pale lilac background, a cube, a tall cylinder, a flat dish and an upright ring, linked by three thin violet arcs.

Your provider connects the phone number. Your provider handles the phone network connection. Your provider handles the remaining transfer flow. The guide names Twilio as an example SIP trunking provider and tells you to purchase a number from that provider before anything else happens.

That is not a footnote. It is the architecture.

The outbound flow is a POST to the Live sessions endpoint with the transport type set to sip. The request carries the destination phone number, a provider endpoint that must use a secure sips scheme for TLS signaling, digest authentication with your provider’s SIP username and password, and a caller number. OpenAI’s own note is blunt about where the credentials live: keep your API key and SIP credentials on your server.

A few specifics decide whether a build survives contact with a real list:

  • The destination must be a full phone number. The guide requires international format in transport.destination and states that SIP URI destinations aren’t supported.
  • Outbound runs on the Live API. The doc says outbound SIP calling is available to all organizations through the Live API, not the Realtime API call-creation endpoint. Pick the wrong one and the call never exists.
  • Your trunk has to support the right transport. TLS signaling and SDES-SRTP media, with SRTP enabled in the provider’s settings before placing a call.
  • The provider endpoint is strict. No credentials, paths, or extra parameters in the URL, and private or local addresses are rejected outright.

Now add what the trunk provider needs on its side. Twilio’s Elastic SIP Trunking documentation asks for a Twilio account, a SIP-enabled network element such as a session border controller, SIP call server, or IP-PBX with internet access, and enough bandwidth for peak traffic. It even gives the math: peak trunk bandwidth equals max simultaneous calls times 100 kbps, which covers the G711 codec plus overhead.

Read that list again and ask who on your team owns it. That is the real answer to the question.

The Caller ID on an AI Phone Call Is Still Your Number

This is the field most buying conversations skip, and it is the one that decides pickup rate.

The outbound request includes transport.trunk.caller_number, the caller phone number sent to your provider. OpenAI does not hand you a number. It sends whichever number you supply.

So the AI is not calling from some neutral machine identity. It is calling from your number, with your number’s history, and your number’s reputation.

Think about what that means on a list of a few thousand records. The model can be flawless and the call still goes nowhere, because the handset showed “Spam Risk” and the buyer let it ring out. No prompt fixes a flagged number. That is a telephony problem with a telephony fix, which is why the number a buyer actually sees belongs to the dialing platform rather than the model. Kixie keeps those as separate pieces, with local presence dialing covering the displayed number and ConnectionBoost sold as a premium add-on aimed at connection quality and dropped calls.

The inbound direction has a matching trap, and OpenAI states it directly. Treat the SIP headers on an incoming call as untrusted caller metadata, not authorization.

Read that as a rule rather than a technicality. The number on the screen is a claim, not an identity. Anything downstream that unlocks an account, confirms a balance, or skips a verification step because of the caller ID is building on a claim the docs already told you not to trust.

Call Limits That Break a Real Calling Block

Published limits are the fastest way to find out whether a demo will survive a Tuesday.

Three frosted glass vessels of different heights on a pale lilac background, each holding a different level of violet liquid, with a frosted glass bar floating above them.

OpenAI lists three that are not negotiable. Ringing is capped at 3 minutes. A connected call is capped at 2 hours. Both are stated as not configurable in the creation request.

Then there is the one that quietly breaks monitoring. Your backend attaches to a sideband connection to watch call progress, and that sideband replays events from only the preceding 3 seconds before switching to live events. Attach late and the ringing and answered events are simply gone.

Why does a 3 second window matter to a sales team? Because call progress is your disposition data.

The sideband is where transport.ringing, transport.answered, and transport.failed arrive. That is how you know whether the call rang, whether a human picked up, or whether setup failed. Miss those events and the call outcome never lands anywhere a manager can inspect. The rep has no attempt history, the CRM has no disposition, and the record looks untouched.

Errors have their own shapes too. Invalid configuration comes back on the creation request, transport setup failures can return a 502, and initialization timeouts return a 504. After the session is created, asynchronous failures only surface on the transport failed event.

So the checklist is not “did the AI sound good.” It is: did the call connect, did the event stream get attached in time, and did the outcome reach the system of record.

Who Answers When the AI Phone Call Needs a Human

Every serious outbound build hits the moment the model should stop talking.

OpenAI supports it. A transfer is a POST to the session refer endpoint with a target, which becomes the SIP Refer-To header and can be a telephone URI or a SIP URI. A hangup is a POST to the hangup endpoint with no body. Both return an empty success response.

And then the familiar sentence again: your provider handles the remaining transfer flow.

Keypad input works in both directions as well. The sideband receives a DTMF event when the caller presses a key, and your backend can send a tone to the trunk. That is how the AI navigates a phone tree, and how a caller picks an option.

Here is the part worth arguing about. A transfer is not a feature, it is a staffing decision. Transferring to a rep who is not at a desk is worse than never offering the transfer, because the buyer already agreed to talk and then got dropped. Decide the hours, the queue, and the fallback before the first dial.

Ask the uncomfortable version. What happens to that buyer at 6:40pm on a Friday?

If the honest answer is a voicemail box nobody checks until Monday, the transfer is decoration.

Answering an Inbound Phone Call Is the Easier Half

Most of the risk above belongs to outbound. Inbound is a different shape, and usually a better first build.

The flow starts at your provider. You purchase a number, then point the provider’s SIP trunk at a project-specific OpenAI address, and subscribe a webhook endpoint to the incoming transport event. When a call arrives, OpenAI sends that webhook and waits.

Then your backend decides. The docs are explicit that you apply your own authorization and routing rules, and either accept the call or reject it.

That accept-or-reject moment is the whole reason inbound is easier to govern. Nobody was dialed without consent, because the buyer called you. There is no calling window to respect and no caller ID reputation to defend. The number rang because someone wanted it to.

There is a deployment detail worth flagging early, because it is painful to retrofit. OpenAI publishes regional SIP endpoints, including one for European data residency and one for SIP hosts physically located in India. If call audio has to stay in a region, that is a routing decision made at setup, not a setting toggled later.

The docs also warn about a duplicate-handler failure that is easy to ship. Webhook deliveries can repeat, the same pending call can surface on more than one connection, and the fix is one handler per action so a tool never fires twice. Translated to a sales floor: one answer path, or the same lead gets two confirmations and one angry reply.

Consent and Recording Rules Do Not Change Because AI Made the Call

Short section, because the rules are the same rules.

An outbound sales call is an outbound sales call. Who or what generates the audio does not reset consent, calling windows, identification requirements, do-not-call obligations, or recording law. If anything, an automated voice narrows what you may do without permission.

We covered that ground separately in TCPA and AI calling for sales teams, and the same logic applies to texting in ChatGPT SMS bot costs and consent.

The operational point is simpler than the legal one. Consent state, opt-outs, and calling windows have to live where the dialing happens, because that is the only layer that can refuse to place the call. A model that has already been handed a phone number is downstream of the decision.

How Sales Teams Actually Use ChatGPT With Phone Calls

Most teams asking this question do not want an autonomous AI caller. They want the admin work gone.

Kixie is direct about the starting point: it does not have a native ChatGPT integration. Teams connect the two with custom builds on the Kixie API and webhooks, or through Zapier, Make, or n8n. That is worth knowing before anybody promises a one-click setup.

What the connection actually moves is call activity. Kixie webhook events cover Start Call, Answered Call, End Call, Voicemail, SMS, Disposition, Call Outcome, Scheduled Activity, and a Conversation Intelligence summary, so what leaves the platform is the record of what happened on the phone.

The other direction is narrower than it looks, and this is where builds go wrong.

Kixie’s developer API has no direct inbound auto-call endpoint. Placing a call or sending a text from an outside trigger runs through Zapier actions instead, which cover making a call to an agent or queue, sending an SMS, and adding or removing a contact from a PowerList or Cadence. Anyone promising a native auto-call API is describing something that does not exist.

That shape is the boring version, and it is the one that works. The rep makes the call on the PowerDialer, which dials up to four leads at a time. The conversation generates a recording and an outcome. Then the model turns that into a summary, a next step, a follow-up draft, and a CRM note.

Notice what the model is not doing. It is not choosing the number, not holding the consent record, not deciding the calling window, and not carrying the audio.

Run the evaluation that way and the question changes usefully. Not “can ChatGPT call my leads,” but “what does the rep stop doing by hand, where does the outcome get logged, and what can a manager inspect on Monday.” The ChatGPT and Kixie integration page lays out the build patterns teams use for exactly that.

If the answer to those three questions is vague, the AI is not the missing piece. The workflow is.

What to Inspect Before You Let AI Make Phone Calls

Treat this like any other tooling decision. The question is never whether the demo sounded good.

Start with ownership. Somebody has to own the phone numbers, the trunk, and the bandwidth. If no name goes next to those, the project is not staffed.

Then walk one call end to end and check where each thing lands:

  • The number. Which number appears on the handset, who bought it, and what is its spam status today.
  • The consent state. Where the opt-out lives, and which layer is able to refuse to place the call.
  • The calling window. Whether the system checks the time where the buyer is, not where the server is.
  • The event stream. Whether your backend attaches in time to catch ringing, answered, and failed.
  • The disposition. Which CRM field holds the outcome, and whether a manager can filter on it.
  • The transfer path. Who picks up, during which hours, and what happens outside them.
  • The recording. Where the audio is stored, who can play it, and under which rules.

Nothing on that list is a model capability. Every one of them is a telephony and process capability.

That is the honest summary of this whole question. The model is the easy part now.

Pick the workflow you want to remove from a rep’s day. Build that one. Then ask whether it needed voice at all, because a large share of the value here is a summary, a logged outcome, and a follow-up that writes itself after a human made the call.

Frequently Asked Questions

Can ChatGPT call a phone number on its own?

Not by itself. The model can carry the conversation on a call, but reaching a phone number requires a SIP trunking provider, a number you purchased, and a session created through the Live API with your trunk credentials and caller number supplied by you.

What is the difference between ChatGPT voice mode and a phone call?

Voice mode is a conversation inside an application, running over WebRTC or a WebSocket. A phone call travels the carrier network, needs a real phone number on both ends, and has to handle ringing, answering, transfers, keypad input, and failure.

Does OpenAI give you a phone number?

No. OpenAI’s telephony guide tells you to purchase a number from a SIP trunking provider such as Twilio, then point that provider’s trunk at your OpenAI project.

How long can an AI phone call last?

Ringing is limited to 3 minutes and a connected call is limited to 2 hours. OpenAI states these limits are not configurable in the creation request.

Can an AI phone call transfer to a live rep?

Yes. A transfer is sent to the session refer endpoint with a target, which becomes the SIP Refer-To header, and the provider handles the rest of the transfer flow. Staffing that transfer is your problem, not the model’s.

Will an AI call get better pickup rates?

No, not on its own. The outbound request sends whichever caller number you supply, so pickup still depends on number reputation and the displayed local number, both of which are handled in the dialing platform rather than by the model.

Does Kixie have a native ChatGPT integration?

No. Kixie states it has no native ChatGPT integration, and teams connect the two through Kixie webhooks, the Kixie API, Zapier, Make, or n8n.

Sources

Sources verified and content reviewed by the Kixie Research Team on November 15, 2026.