Your AI receptionist has to say it is AI. Here is the line, and the receipt.
An AI voice on the phone is treated as an artificial voice by the FCC, and California requires the call to open with a statement that the voice is AI-generated. BotNotice gives you the line, tells you which rules it satisfies, and turns your voice platform's webhook into a receipt for every call.
The line
“Hi, this is Acme Dental. I'm an automated assistant using an AI-generated voice, not a person. This call may be recorded. If you'd rather speak with a person, just say ‘representative’ at any time. How can I help you today?”
Three parts in the first sentence: who, that it is AI and not a person, and that a person is available. Available in English, Spanish, French and German, and editable so it sounds like you.
How it works
- 1Determine
From your profile (which states your callers are in, your industry), BotNotice works out which rules reach your calls and gives you the line the call must open with.
- 2Say it
Paste the line into your voice platform as the first thing the agent says. Retell: begin message. Vapi: first message. Your own code: read the policy endpoint at call start.
- 3Get the receipt
Point the platform's webhook at BotNotice. Each call is recorded as started, disclosed, handed to a person, ended. With a transcript, the disclosure is confirmed and timed.
- 4Keep the proof
The receipt lives in the same ledger as your chat notices, kept 7 years on plans with records, with the notice version in force and a chain check nobody can quietly edit.
Which rules reach a voice agent
- FCC ruling on AI-generated voices (TCPA): AI voices are artificial voices; identify the business at the start; consent and calling-hour rules apply to outbound calls. $500 to $1,500 per call in private suits.
- California AI voice disclosure for automated calls: a statement at the beginning of the call that the voice is artificially generated, plus who is calling.
- State chatbot laws written for “any AI that converses with a person” (Utah, Maine, California SB 243 and others) reach inbound AI calls as well as web chat. Laws by state.
Your dashboard shows the ones that reach your callers, from your profile, with the notice version each call cites. Nothing here is legal advice.
Connecting a platform
Put the line in the agent's begin message. Paste the webhook URL from your dashboard into the agent's webhook field. Retell's call_started and call_ended messages are understood as they are, and the transcript on call_ended confirms the disclosure.
Put the line in the assistant's first message. Set the server URL and the server secret from your dashboard, and turn on status-update and end-of-call-report. The end-of-call report's transcript confirms the disclosure.
Send a small JSON message to the webhook URL when things happen on the call:
{"call_id":"abc123","event":"disclosed"}
{"call_id":"abc123","event":"handoff_human","agent":"Maria"}
{"call_id":"abc123","event":"ended","transcript":"…"}Read the policy at call start and speak what it returns:
GET /api/voice/policy?k=YOUR_SITE_KEY
→ { "disclosure": "Hi, this is …", "timing": "start",
"askForHuman": true, "laws": [...], "policyVersion": 12 }The exact URLs and your token are on your site's Voice page in the dashboard. Full API notes on the integrations page.
Questions
Does an AI phone agent have to say it is AI?
For outbound calls, yes in practice: the FCC treats AI-generated voices as artificial voices under the TCPA, California requires a statement at the start of an automated call that the voice is AI-generated, and several state chatbot laws that cover "any AI that converses with a person" reach inbound calls too. Saying it in the first sentence satisfies all of them at once.
What exactly should the agent say?
Three parts, in the first sentence: who is calling or answering (the business), that the voice is AI and not a person, and that a person is available on request. BotNotice gives you a line per language that does this and lets you edit it.
How does BotNotice know the disclosure was actually spoken?
Your voice platform sends a webhook at the end of each call. If it includes the transcript, BotNotice reads it once, checks that the disclosure appears and whether it fell at the start, writes the result to the record, and discards the transcript. If there is no transcript, the platform can report the disclosure right after the greeting and the record says it was reported, not confirmed.
What is stored about the caller?
Nothing. Only the platform's call id, the times, which notice version was in force, and how the disclosure was confirmed. No phone numbers, no names, no audio, no transcript.
Which platforms work?
Retell and Vapi webhooks are understood as they are. Bland, Twilio ConversationRelay and anything with a webhook or a few lines of code use a small JSON shape: a call id and an event name. The dashboard shows the exact URL and token to paste.
Is a voice disclosure counted like a chat notice?
Yes. Each disclosed call counts toward the website's monthly notices, and on plans with records it is written to the same tamper-evident ledger as chat notices, under the voice channel, with the call id so a complaint can be matched to one call.
Related: how the record works · when you need this · AI disclosure laws by state.