Honoring opt-outs from AI call transcripts and STOP texts
The rule changed in April 2025: a person can revoke consent by any reasonable means, and you have to honor it across every channel. For an AI agent that means reading what the person said on the call.
What the rule says
The FCC's revocation order (effective April 11, 2025) says consent can be revoked "in any reasonable manner": STOP, QUIT, END, REVOKE, OPT OUT, CANCEL and UNSUBSCRIBE by text, and any plain-language request on a call, by email or on a form. You cannot require a specific channel or a specific word. A revocation must be honored within ten business days, and it applies to the channel it came in on and, when the words are general, to every channel. A STOP reply to a text must get at most one confirmation message and nothing else. The "revoke all" provision, under which a revocation to one business covers every business the caller represents, has been delayed to January 31, 2027, but treating it that way now is the safe default for an agency.
Why AI agents get this wrong
A human rep who hears "take me off your list" writes it down. An AI agent hears it, says something polite, and the conversation ends; nothing is recorded unless the transcript is read afterwards by something that knows what to look for. Most teams don't read transcripts for revocations, so the next campaign calls the same person again, and now the violation is willful: $1,500 per call instead of $500, with the transcript as the plaintiff's evidence.
What to do
- Point your platform's end-of-call webhook at something that reads the person's words. Retell sends
call_endedandcall_analyzedwith the transcript and word timings; Vapi sends anend-of-call-report. Only the customer's turns count; ignore what the agent said. - Treat clear requests as revocations immediately: "stop calling me", "take me off your list", "don't call this number again", "remove me". Apply them account-wide, on every channel, within seconds, not days.
- Flag unclear ones for a person: "not interested", "wrong number", "I'm busy", a bare "stop" that might mean stop talking. A human decides in a minute; the agent should not.
- Do the same for texts. STOP, UNSUBSCRIBE and friends from Twilio or Telnyx inbound webhooks apply instantly; send one confirmation and nothing more.
- Keep the evidence: what the person said, when (to the second, from the transcript timings), when it was applied, and every attempt blocked since. If they are called again by mistake, you want to show it was blocked, not argued about.
How Call Clearance does it. It accepts opt-outs from Twilio, Telnyx or plain JSON text webhooks, from Retell and Vapi end-of-call webhooks, from GoHighLevel, and from the API. Clear requests are applied within seconds for every business on the account and every channel; unclear ones are flagged on the dashboard for a person. Each one keeps the person's words, the time, the latency to apply, and the attempts blocked since, hashed into the same chain as every check. The Call Record for a blocked call shows the opt-out that blocked it.
Words that count
Clear, apply at once: stop, stop calling, stop texting, unsubscribe, remove me, take me off, do not call, don't call again, opt out, cancel, quit, end. Unclear, flag for review: not interested, wrong number, busy, call back later, who is this. A person saying "stop" mid-sentence to interrupt the agent is the hardest case; the timing and what follows usually settle it, and a human should look.
Check your first number in five minutes.
Free plan, no card. Opt-outs, consent and calling hours run from the first call; the national list switches on when a business's FTC registration is on file.