Automate only messages that pass a gate: written consent for that channel on file, a do-not-call check, a send time inside the lead's calling window, working opt-out handling, and a licensed human for anything that sounds like advice. The TCPA sets the calling and texting limits, so build the gate into your CRM before any AI sends a word.
Speed and compliance only clash when consent is missing. A lead who ticked a proper consent box can be messaged within a minute. A purchased lead with no consent record cannot be texted at all, however fast the automation is. The TCPA's damages are $500 per call or text, and a court may raise that to three times as much for wilful violations, so the gate has to sit in front of the AI rather than behind it.
What the rules say, in plain terms
These are the rules that matter most to an agency following up quote requests, read from the primary texts and regulators' pages in November 2026. They come from the Telephone Consumer Protection Act (TCPA), the calling rules made under it, and the telemarketing and commercial-email rules that sit beside it. If you reach leads in other markets, their rules differ, and none of this is legal advice: your lawyer or compliance contact approves the final settings.
| Topic | What the rule says |
|---|---|
| Consent for marketing by autodialler or artificial voice | "Prior express written consent" is a signed written agreement that clearly authorises the seller to send advertisements or telemarketing by automated dialling or an artificial or prerecorded voice, includes the number, and tells the signer clearly that agreeing is not a condition of buying anything |
| AI-generated voices | A February 2024 ruling confirmed that the TCPA's "artificial or prerecorded voice" covers current AI technologies that generate human voices, so such calls need the called party's prior express consent, and written consent if they carry an advertisement or telemarketing |
| Calling hours | No telephone solicitation to a residential subscriber before 8 a.m. or after 9 p.m., measured as local time where the called person is |
| National Do Not Call Registry | Registered numbers cannot be called for sales unless an exemption applies: an existing relationship (up to 18 months after a purchase, delivery or payment, or three months after an inquiry or application) or the person's signed written agreement to calls from or for the seller. Callers must sync with the registry at least every 31 days |
| Internal do-not-call list | Keep a list of people who ask not to be called; honour each request within a reasonable time that cannot exceed ten business days; keep honouring it for five years |
| Revoking consent | From 25 November 2026 a caller may name one exclusive opt-out method (a key-press option on the call, a text reply using "stop", "quit", "end", "revoke", "opt out", "cancel" or "unsubscribe", or a named website or phone number) if it discloses that method clearly in each call or text; a caller that names none must accept any reasonable method. A stop in reply to a marketing message ends consent for all of that caller's marketing calls and texts. Requests must be honoured within ten business days |
| Accurate headers, a clear sign that it is an advertisement, your postal address and a free opt-out honoured within ten business days; business-to-business email is covered | |
| Penalties | TCPA: $500 per violation, up to three times that if wilful or knowing. The telemarketing and CAN-SPAM guidance pages listed civil penalties of up to $53,088 per violation and per email in November 2026 |
Three clarifications stop the common misreadings. The existing-relationship exemption lets a person call a registered number; it does not replace the written consent needed for automated texts or artificial-voice calls. Marketing texts are treated as calls under these rules. And whether your texting software counts as an autodialler has been argued over since a 2021 top-court ruling narrowed the definition, which is a legal opinion rather than a setting, and some states use broader definitions. One state's statute requires written consent for automated sales calls and sets damages from $500 a call, and a companion section bars sales calls after 8 p.m. and more than three calls on the same subject in 24 hours. So design for the strict reading and let counsel loosen it.
Consent you can prove: the form wording and the fields behind it
Consent that exists only in someone's memory is worth nothing in a dispute. The agency has to show what the person saw, when, and what they agreed to. Start with the wording on your own quote form. This is an illustrative block for your lawyer to edit, not finished legal text:
[ ] By selecting "Get my quote" I agree that [Agency name] may contact me about insurance
quotes by phone call, text message and email at the number and address I entered, including
by automated dialling or messaging technology. I understand that I am not required to agree
as a condition of buying anything. Message and data rates may apply. Reply STOP at any time
to opt out.It names the seller, covers the channels, mentions automation, repeats that consent is not a condition of purchase and gives the opt-out. If you plan to use AI-generated voice calls, the wording has to say so, because the 2024 ruling treats those as artificial-voice calls. Keep the box unticked by default, and keep it separate from the terms of use so nobody can argue it was buried.
Wording is half the evidence. The other half is a record created at the moment of submission, stored on the lead itself:
| Field to add to the lead record | Why it matters |
|---|---|
| Consent timestamp and time zone | Proves when; also drives the send window |
| Consent text version | Proves exactly what the person agreed to, even after you edit the form |
| Page URL and form name | Proves where it was collected |
| Phone number and email as entered | The written agreement has to cover the number you actually use |
| Channels agreed | Separate flags for call, text and email, so the gate can check each |
| Seller named | Confirms your agency, not "partners", is the party authorised |
| Source and vendor | Own form or purchased; if purchased, the vendor and lead ID |
| Registry check date | Shows the number was checked within the last 31 days |
| Opt-out date and channel | Drives suppression and proves you honoured it |
Third-party evidence helps where you cannot control the form. ActiveProspect's TrustedForm page describes a certificate for each lead showing a time stamp, session replay, event log, page URL and the consent language, shareable by web address and kept for up to five years, with pricing that scales with volume. Whatever tool you use, the test is the same: could you hand the record to your lawyer and have them see what the person saw?
The common ways agencies fail this section are dull ones. A pre-ticked box. One tick that bundles "send me a quote" with "contact me about anything". A link to a list of dozens of "marketing partners" in place of naming the agency. Consent captured on an old version of the form that nobody archived. Fixing the form takes an afternoon, and none of those problems can be fixed retroactively after a message has gone out.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThe pre-send gate: eight checks before any automated message goes out
The gate is a set of plain conditions your CRM or automation platform checks before every send. If your platform cannot hold a message until a condition is met, that is a reason to switch platforms, not to skip the check. Run the cheapest checks first and log every decision with a timestamp.
| # | Check | Data it reads | If it fails |
|---|---|---|---|
| 1 | Consent exists for this channel | Consent timestamp, text version, channel flags | Route to a human; no call or text |
| 2 | The agency is the named seller | Seller field in the stored consent text | Treat as no consent for call and text |
| 3 | Registry and relationship status | Registry check under 31 days old; inquiry date or last purchase date | No live call unless an exemption applies; no automated text either way without written consent |
| 4 | Not suppressed | Internal do-not-contact list across every channel | Stop; log the reason |
| 5 | Inside the send window | Lead's time zone, taken from the address on the form | Queue to the next window; send an email now if email consent exists |
| 6 | Under the frequency cap | Touch count by channel in the last 24 hours and 7 days | Hold; a producer decides whether to continue |
| 7 | Message is an approved template | Template ID and approval date | Route to human review; no free-written messages |
| 8 | A licensed person is available for this lead | Lead's state against each producer's licence list | Offer the callback only when someone can take it; otherwise decline politely |
Three details make the gate work in practice. For check 5, the rule is about local time where the person is, so use a time zone from the address on the form; an area code on a mobile number says where the number was issued, not where the owner is today. I would also shrink the window to 9 a.m. to 8 p.m. in the lead's zone, because it leaves margin and is closer to the tighter windows some states set. For check 6, pick a cap with your lawyer, starting from the strictest state you serve. For illustration, three attempts in the first day and one a day after that is a conservative opening position that you adjust rather than copy. For check 8, licensing rules generally limit who may discuss coverage and give advice, so let the automation arrange a callback and never answer a coverage question itself.
The gate also gives you a better speed-to-lead measure than raw reply time. Track time to first compliant contact: the minutes from the quote request to the first message or call that passed all eight checks. A lead submitted at 22:10 will not be contacted by phone until the next morning. An immediate email, where the email consent exists and the message carries your postal address and an opt-out, keeps the lead warm without breaking the calling window. The tutorial on replying to every enquiry in under five minutes covers the mechanics of fast replies; the gate decides which of them you are allowed to send.
The CRM fields in the earlier table belong in whichever system you already use. The tutorial on HubSpot for insurance agents shows where such properties and workflow conditions sit in a typical agency pipeline, and the idea transfers to any CRM that supports conditional workflows.
Opt-outs: design for the strictest reading
The rules allow up to ten business days to honour a stop request. Treat that as a legal ceiling, not a service level: an automation that keeps messaging someone for a week after they typed STOP is the kind of record that ends up in a complaint. Suppress within the hour, on every channel, and write the date and the words the person used.
The scope of a stop request was settled in part in October 2026. Under amendments that apply from 25 November 2026, a stop in reply to a marketing message covers all of that caller's marketing calls and texts, while a stop in reply to a purely informational message, such as an appointment reminder, may be limited to that kind of message. A companion proposal asks whether to shorten the ten-business-day limit and whether to require a revoke-all option, so the rules may tighten again. Building to the strict reading costs little: treat any opt-out as a stop for all automated marketing from the agency, on every channel, until a person re-consents, and recognise all seven standard reply words even if you name a different opt-out method. One confirmation text is allowed if it only confirms the opt-out and carries no marketing; send it within five minutes, since a later one needs a reason for the delay. If someone replies to a quote text with "stop sending me auto quotes, I still want the home one", suppress everything and have a licensed producer email them to confirm what they want. A lost follow-up is cheaper than a contested one.
AI is useful for sorting replies, because real replies are messy. This prompt reads one reply at a time and returns a label your automation can act on, with a human catching the ambiguous ones:
Classify this reply to an insurance quote follow-up. Return one label and a one-line reason.
Labels: OPT_OUT (any wish to stop, however phrased), WANTS_CALL, WANTS_QUOTE_INFO,
COVERAGE_QUESTION (needs a licensed person), COMPLAINT, WRONG_NUMBER, UNCLEAR.
If the reply could be read as an opt-out in any way, choose OPT_OUT. Do not answer the reply.
Reply: "{{reply_text}}"| Reply (illustrative) | Label | What the automation does |
|---|---|---|
| "STOP" | OPT_OUT | Suppress all channels, log date and words |
| "not interested thanks, take me off" | OPT_OUT | Suppress all channels; no confirmation chase |
| "can you call after 6?" | WANTS_CALL | Create a task for a producer; book inside the window |
| "does this cover my son as a named driver?" | COVERAGE_QUESTION | Task for a licensed producer; no automated answer |
| "wrong number" | WRONG_NUMBER | Suppress the number and flag the lead record as bad data |
| "ok" | UNCLEAR | Human reads the thread before anything else is sent |
Run the classifier as a second line only. Keep a plain keyword rule ahead of it that suppresses on "stop", "quit", "end", "revoke", "opt out", "cancel" and "unsubscribe", because those are the standard opt-out words the rules list and should never wait on a model's judgement.
What AI drafts and what a licensed human decides
| Task | AI | Human |
|---|---|---|
| Follow-up templates (confirmation, reminder, document request) | Drafts and varies the wording | A licensed producer approves each template once, with its date recorded |
| Filling in merge fields | Not needed; the CRM does it | Checks that fields never carry health or claims details |
| Sorting replies | Labels them, with opt-out failing safe | Reads UNCLEAR and COMPLAINT replies the same day |
| Summarising a call for the file | Drafts notes from a recording the caller was told about | Corrects and signs off |
| Coverage questions, quotes, recommendations | Nothing | A licensed person, always |
| Voice calls to leads | Not for marketing without written consent that covers artificial voice | A person dials |
Approving the template, not each message, is what makes this affordable. Drafting a set looks like this:
Draft 3 follow-up texts and 1 email for a person who asked for an auto insurance quote on our website.
Rules: name the agency ([Agency name]); one plain sentence of purpose; offer a call time or a
reply; no prices, no coverage statements, no "lowest rate" claims, no urgency or pressure;
texts under 160 characters using straight apostrophes only; end every text with "Reply STOP to opt out";
the email needs an opt-out line and a [postal address] placeholder.A typical first draft (illustrative) has a text such as "Hi [first name], it's [Agency name] about your quote request. Want a call today or tomorrow? Reply STOP to opt out." Look for four faults before approving it. A phrase like "you're covered" or "you'll save" has crept into a line, which is advice and a claim. The curly apostrophe in "it's" can push a text into the 70-character encoding, splitting a 160-character message into several billed parts, which is why the prompt asks for straight apostrophes. The email may lack the postal address. And an "act now" line has appeared where you asked for none. Fix the template, record who approved it and when, and only then load it into the automation.
Before you buy leads: seven questions for the vendor
Most pages on this keyword push lead buying. A bought lead is the riskiest input to an automated follow-up, because the consent was collected by someone else on a form you never saw, and the liability for the call or text is still yours. Get answers in writing before the first lead arrives:
- Is there evidence for every lead? A certificate, session record or screenshot with timestamp, page URL and consent text, delivered with the lead and not on request after a complaint.
- Does the consent text name your agency? The written agreement has to clearly authorise the seller. A link to a long list of "partners" is weak, and you should check your name is on it.
- Does it cover the channel you will use? Consent to "be contacted" is not clearly consent to texts or to artificial voice.
- How many agencies receive each lead? Shared leads are called by several firms in the same hour, which raises complaints and opt-outs.
- When was the number checked against the registry? A date older than 31 days is a flag.
- What happens to leads with no valid consent? You want replacement or a refund, and a contract clause that returns the cost of any complaint traced to a vendor's consent record.
- Can you see the form? Look at the live page and the archived versions yourself.
The rule that would have required consent to name only one seller was struck down by an appeals court in January 2025, and the decision left the older standard in place; check for any later replacement before you rely on this. The older standard still asks for a written agreement that clearly authorises the seller, so the question about your name matters either way. If a vendor cannot supply evidence for a lead, treat that lead as one you may, subject to your lawyer's view of the email rules above, email or write to, and keep it out of the call and text path.
400 leads a month through the gate: a worked example
An agency with six licensed producers receives about 400 leads a month: 150 from its own quote form and 250 bought as shared leads. The figures are illustrative. Before the gate, every lead went to the same text-then-call sequence. After the gate, the sorting looks like this:
| Source | Leads | Result at the gate | First-touch path |
|---|---|---|---|
| Own quote form | 150 | 138 have complete consent records. 12 fail: 5 came through an old form with a pre-ticked box, 4 lack a consent-text version, 3 are on the internal do-not-contact list | 138 text and call. The 12 get email only after a producer reviews them (the 3 suppressed leads get nothing) |
| Bought shared leads | 250 | 90 arrive with evidence naming the agency. 160 have none | 90 text and call. 160 email or mail only |
| Total | 400 | 228 pass for call and text (57%) | 172 are not eligible for call or text (43%): 169 email or mail only, 3 no contact |
The sum that justified the work: one automated text to each of the 160 leads without evidence is 160 times $500, or $80,000 of statutory exposure from a single message, before a second text or a call. That is a ceiling on what a claimant could ask for, not a prediction, but it shows why the gate's cost is small beside what it prevents.
Four leads show how the rules interact. Lead one submits at 19:05 with full consent and gets a templated confirmation text within a minute and a producer call at 19:20. Lead two submits at 22:10 with consent; the gate holds the text and call until 09:00 in the lead's zone, and an immediate email goes out because email consent exists. Lead three is a bought lead with no evidence; the gate sends nothing by phone and queues one email, and the vendor is asked for the missing certificate. Lead four replies "unsubscribe" to the email and "STOP" to an earlier text; the keyword rule suppresses both channels within the hour, and the classifier is never consulted.
Of the 38 own-form leads that arrived outside the window in the month, all received an email within a minute and a call the next morning. The agency's tracking metric changed from "average reply time" to "time to first compliant contact", split by source. Twelve AI-drafted templates were approved by one licensed producer in about 90 minutes, and 61 replies arrived in the month: the classifier handled 55 cleanly, 6 went to a person as unclear, and 9 opt-outs were processed in the same hour they arrived.
Records and reviews that keep the gate honest
A gate that nobody audits drifts. Keep consent records, gate decisions and opt-outs for at least five years, which matches how long the calling rules say a do-not-call request must be honoured and how long TrustedForm certificates are kept; ask your lawyer whether your own limit is longer. Then put three reviews in the calendar:
- Monthly: pull 20 sent messages at random and trace each back to its consent record, send time and approved template. Any message you cannot trace is a finding, whatever the cause.
- Quarterly: have the licensed producer and your lawyer re-read the templates and the consent wording, and compare opt-out timings with the suppression log.
- When the rules move: the opt-out amendments apply from 25 November 2026, and the proposal to shorten the ten-business-day limit is still open, so re-read the gate's settings when it is decided rather than waiting for the next quarter.
Rules move quickly in this area, which is why the settings in the gate (window, cap, suppression scope) sit in one place and are not scattered through twenty workflows. Changing one number should change every send.
Two neighbouring jobs deserve their own care. Letting AI answer calls and qualify leads raises the same consent and disclosure questions on the inbound side; the tutorial on AI answering calls and qualifying leads for an insurance agency covers it. If you are tempted by AI voice for outbound follow-up, read AI voice agent follow-up and the TCPA first, since the 2024 ruling on AI voices applies to every sector. And for the broader question of recording consent on your website, see the tutorial on AI chatbot consent management for a professional firm's site.
The main point holds across all of it. AI drafts, sorts and schedules; the rules decide whether a message may go; a licensed person answers anything that sounds like advice. Fast follow-up is a good goal, and the agency that reaches a lead first, with a message it was entitled to send, is the one that keeps the lead and its licence.
Insurance lead follow-up questions
Does a quote request give me consent to text the lead?
Not by itself. Automated marketing texts and artificial-voice calls need prior express written consent: a signed agreement that clearly authorises your agency and covers the channel. A quote form with proper, specific consent wording can supply that. A request with no consent language, or consent naming only unnamed partners, may not. Ask your lawyer to approve the wording.
Can an AI voice agent call my leads?
Not for marketing without prior express written consent that covers artificial voice. A February 2024 ruling confirmed that AI-generated voices count as artificial or prerecorded voices under the TCPA, so the same consent, identification and opt-out duties apply. Many agencies keep AI voice for inbound calls and have a person dial outbound.
How fast must I honour a stop request?
The rules allow up to ten business days, and a pending proposal asks whether to shorten that, so treat it as a ceiling. Suppress within the hour on every channel, record the date and the words used, and always honour the standard reply words (stop, quit, end, revoke, opt out, cancel, unsubscribe), which the rules list as a recognised way to opt out.
Are purchased insurance leads safe to text?
Only if you hold evidence of consent that clearly names your agency and covers texting. The liability for a text or call stays with you even when a vendor collected the consent. Ask for a certificate or record per lead, read the consent wording, and keep any lead without evidence out of the call and text path.
What can AI write in insurance follow-up messages?
Logistics: confirmations, reminders, document requests and callback offers, drawn from templates a licensed producer has approved. It should not quote prices, describe coverage, recommend products or add pressure. Anything that sounds like advice goes to a licensed person, and free-written messages should not reach leads without review.
Further reads
- AI Real Estate Cold Calling: Scripts, Lists and Do-Not-Call Rules — Do-not-call rules and call scripts for another lead-driven profession.
- How Much Does It Cost to Automate Lead Follow-Up With AI? — What lead follow-up automation costs to set up and run.
- AI Answering Service for Insurance Agents: Claims Calls and Quote Leads — An AI answering service for claims calls and quote leads.
- Best AI Tools for Insurance Agents: Seven for Quotes, Renewals and Calls — Seven AI tools for quotes, renewals and calls.
- How to Automate Insurance Renewals With AI and Keep More Clients — Automate renewals so existing clients stay without chasing.
- AI Lead Generation for Financial Advisers Without Cold Outreach — Compliant lead generation for financial advisers without cold outreach.
- AI Insurance Agency Management System: What to Ask Before You Switch — What EZLynx and Vertafore say their AI does, how to score AI and data export together, and the questions to put to any agency management vendor.
- Zoho CRM for Insurance Agents: Policies, Renewals and Zia — A step-by-step Policies module and renewal rules in Zoho CRM, the edition limits that matter, and what Zia adds for an insurance agency.
- AI Chatbot for an Insurance Agency: Quote Requests, Not Advice — Quote-request-only scope for an agency website bot: fields to collect, replies that refuse, and a hand-off packet for the licensed agent.
- AI Tools for Life Insurance Agents: Follow-Up Within the Rules — A rule-first method for AI follow-up in life insurance: guard-railed prompts, replacement flags, recording consent and a pre-send gate.
- Does Agency E&O Cover AI Mistakes? What to Ask Your Carrier — A per-tool list of questions for your E&O carrier, the exclusion wording to look for, and the written AI policy that gives you evidence. Not legal advice.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: the calling rules at 47 CFR 64.1200 (current text, plus the opt-out amendments published 26 October 2026 and the companion proposal); the February 2024 declaratory ruling on AI-generated voices; the Do Not Call Q&A for telemarketers and the CAN-SPAM compliance guide; the TCPA's private right of action; the January 2025 appeals-court opinion on one-to-one consent; one state telephone-solicitation statute; ActiveProspect TrustedForm pages (all read November 2026).