The AI voice that calls your customers is a robocall
The AI agent that has answered the clinic's phone for a month can make calls too, from Tuesday. Tomorrow's appointments get a spoken reminder, last week's no-shows a call-back, unpaid invoices a nudge in the same warm voice. The setup screen asked for a list of numbers and an agent, plus the hours it may dial. Whether anyone on the list had agreed to be called by a machine was a question for whoever uploaded the list.
Under the Telephone Consumer Protection Act, the TCPA, an AI voice call to a customer is a robocall, and a robocall needs the customer's consent before it is placed. This article is about the Federal Communications Commission ruling of February 2024 that confirmed it, and the rules it points to; states add their own telephone laws, which this piece leaves alone. Every document quoted was read on 12 September 2026. I am not a lawyer, and none of this is advice about your calling list.
The ruling turns on the generated voice, and good intentions do not change it
The sentence that matters to a small business is in paragraph 5 of the ruling:
"callers must obtain prior express consent from the called party before making a call that utilizes artificial or prerecorded voice simulated or generated through AI technology."
FCC, Declaratory Ruling FCC 24-17, CG Docket No. 23-362, adopted 2 February 2024, released 8 February 2024, paragraph 5
The FCC's verb is "confirm": the TCPA has restricted "artificial or prerecorded voice" calls since 1991, a generated voice is artificial because, in the FCC's words, "a person is not speaking", and the ruling's first footnote counts any call made with such a voice as a robocall.
In paragraph 7 the friendly-reminder defence is dealt with directly: "we understand that not all AI-generated calls may be deceptive or annoy all consumers. But the TCPA's demands fully apply to those calls and, thus, consumers can themselves choose whether to receive them."
The ruling covers outbound calls, and leaves the answering side of the same agent alone
Paragraph 9 of the ruling ends: "These requirements are applicable to any AI technology that initiates any outbound telephone call using an artificial or prerecorded voice to consumers." The FCC's notice of proposed rulemaking of August 2024 drew the other edge: "The TCPA's requirements do not extend to technologies used to answer inbound calls."
So the month in which the agent only picked up the phone sat outside the ruling, and consent enters the moment the same agent dials out. The choice of an AI receptionist or a person for the answering side is made on cost and coverage.
Consent comes in two grades, and the reminder needs the cheaper one
For a call to a mobile that sells nothing, the rules require "prior express consent", and the FCC has said since 1992 that handing over your number is a form of it. In its words, "persons who knowingly release their phone numbers have in effect given their invitation or permission to be called at the number which they have given, absent instructions to the contrary." A 2015 ruling of the same body adds the limit: "the call must be closely related to the purpose for which the telephone number was originally provided."
Type your mobile into a clinic's booking form and, on that reading, you have consented to a reminder about the booking; the consent lives in the booking record, in how the number arrived, and the dialling tool imports the number without the record it came from.
The second grade is for calls that "include or introduce an advertisement or constitute telemarketing". Those need "prior express written consent", defined as "an agreement, in writing, bearing the signature of the person called". Add "and we have 20% off whitening this month" to the end of the reminder and, on the FCC's rules, it has become telemarketing, with the written-consent grade that goes with it.
Healthcare providers have an exemption for calls to mobiles. Reminders, confirmations and lab results need no consent if, among other conditions, the number was "provided by the patient", the provider is named at the start and the call is "generally one minute or less". The same rule says such calls "may not include accounting, billing, debt-collection, or other financial content". The exemption describes the clinic's appointment reminder and, by its last clause, excludes the unpaid-invoice nudge sent in the same voice. (Whether spoken reminders reduce no-shows at all is a separate question.) A process audit writes down, before anything dials, which messages sell something and which numbers came with what permission.
The first seconds of the call are written into the rules
Every artificial-voice message must "state clearly the identity of the business" at the beginning, under its registered name, and "state clearly the telephone number" of that business during or after it; a request to stop must be honoured within "ten business days".
The disclosure rule most people expect is still a proposal: the August 2024 notice proposed "requiring callers using AI-generated voice to, at the beginning of each call, clearly disclose to the called party that the call is using AI-generated technology", and as of 12 September 2026 the Federal Register shows no final rule in that docket.
The tool asks for a phone number and an agent
I read the outbound-call documentation of five voice platforms on 12 September 2026: what each setup requires, and where the word "consent" appears.
| Platform | What the outbound call needs | Where consent appears |
|---|---|---|
| Vapi | "Customer's Phone Number", an assistant | A TCPA Consent Guide linked from the outbound page |
| Retell AI | a number, an agent, an API key; batch calls from a CSV for "appointment reminders, payment follow-ups, lead qualification" | Nowhere on the outbound-call or batch-call pages |
| Bland AI | phone_number in international format (E.164) |
Optional in-call "guard rails" such as tcpa:ai_disclosure and tcpa:opt_out; nothing on either page about consent before the call is placed |
| Synthflow | "the agent that does the talking", "the caller ID it dials from", "the list of people to call" | Nowhere on the campaigns page |
| Twilio | to, from and a script in Twilio's own markup (TwiML) |
In the Acceptable Use Policy, as a duty to obey laws that "require (a) consent be obtained prior to transmitting" |
Bland's tcpa:ai_disclosure covers the disclosure that the FCC's August
2024 notice proposed; on the read date neither the Send Call reference nor
the guard-rails tutorial asks whether the number may be called at all.
The reading most people in this market arrive at is that AI phone calls are fine as long as you are not selling something, and that selling needs express written consent. The second half matches the rules; the first half is where paragraph 7 of the ruling bites, because the reminder that sells nothing still needs consent of the cheaper grade. The statute prices the gap at "$500 in damages for each such violation", and a court may raise that to "not more than 3 times" the amount if the violation was wilful or knowing. A list of 300 reminders to numbers with no consent on record is 300 violations.
Where this leaves you
The calls the voice initiates. Reminders, follow-ups, payment nudges and call-backs are all outbound calls under the rules above; if the platform lists them on one screen with the answering side, the screen hides a legal line running through the middle of it.
The place where consent is recorded, per number. For an informational call, "in the booking, where the customer gave it" can be the answer, as long as the call is about that booking; a call that sells needs a signed agreement on file. The register of what is automated is where a "consent source" column belongs, next to the script the voice reads.
If you would rather have someone map which of your automated messages go out, in what voice, on whose permission: that is part of the process audit. $299, three business days, and a written map you keep whether or not you hire us.
Related: when your automation tells a customer something untrue covers who answers for what the voice says once it is on the line.