Prerecorded and AI Voice Robocalls: What the TCPA Prohibits

The rule in one sentence

The TCPA bans calls “using an artificial or prerecorded voice” to cell phones without prior express consent (47 U.S.C. 227(b)(1)(A)), and to home phone lines without prior express consent (227(b)(1)(B)), unless it is an emergency or an FCC exemption applies. If the call is a sales call, the FCC’s rules require prior express written consent for both cell phones and home lines (47 CFR 64.1200(a)(2) and (a)(3)).

No autodialer is needed. After Facebook v. Duguid narrowed the autodialer definition in 2021, recorded-voice claims became a more direct route in many robocall cases, because the only question about the technology is whether a recording or synthetic voice was used.

AI and voice cloning: the FCC’s February 2024 ruling

On February 8, 2024, the FCC released a Declaratory Ruling (FCC 24-17) answering whether AI voices count. They do. The Commission confirmed “that the TCPA’s restrictions on the use of ‘artificial or prerecorded voice’ encompass current AI technologies that resemble human voices and/or generate call content using a prerecorded voice.”

On voice cloning specifically, the ruling says these technologies “are ‘artificial’ voice messages because a person is not speaking them.” It also closed a loophole some callers were testing: the TCPA “does not allow for any carve out of technologies that purport to provide the equivalent of a live agent.” An AI system that holds a conversation with you is still an artificial voice.

The ruling built on the FCC’s 2020 “soundboard” decision. A soundboard call has a live agent clicking prerecorded clips instead of speaking. The FCC held that the presence of a live agent “does not negate the clear statutory prohibition against initiating a call using a prerecorded or artificial voice,” and applied the same logic to AI.

The ruling also cited a Ninth Circuit decision describing an artificial voice as including “a sound resembling a human voice that is originated by artificial intelligence” (Trim v. Reward Zone USA, 2023).

A real example: the New Hampshire deepfake calls

Two days before the January 2024 New Hampshire primary, voters got robocalls with a cloned voice of President Biden telling them to “save your vote for the November election.” In May 2024 the FCC proposed a $6 million fine against political consultant Steve Kramer for “apparent spoofing violations” under the Truth in Caller ID Act. It brought a separate action against Lingo Telecom, the carrier that passed the calls through with the highest level of caller ID attestation. The FCC’s release noted that its AI ruling gives “the FCC and state Attorneys General” the tools to go after callers like these.

Our news coverage of the New Hampshire voice cloning case follows what happened next. A proposed fine is an allegation, not a final finding. The case shows how an AI robocall can arrive: with a spoofed number and a familiar voice.

State attorneys general and AI calls

The TCPA also lets states sue. Under 47 U.S.C. 227(g), a state attorney general who believes a caller is engaged in “a pattern or practice” of unlawful calls to residents can seek an injunction and “$500 in damages for each violation,” trebled for willful or knowing conduct. Before the AI ruling, 26 state attorneys general wrote to the FCC urging that “any type of AI technology that generates a human voice should be considered an ‘artificial voice’ for purposes of the TCPA.” Then-Chairwoman Rosenworcel’s statement on the ruling said it means “State Attorneys General across the country can go after the bad actors behind these robocalls and seek damages under the law.”

For you, this is a second route. If you cannot identify an AI caller well enough to sue, a detailed report to your state attorney general, with the voicemail attached, can feed an investigation that reaches many victims at once.

What a prerecorded message must say

Even a lawful recorded call has to follow the FCC’s identification rules in 64.1200(b):

• At the start: “state clearly the identity of the business, individual, or other entity that is responsible for initiating the call,” using the name registered with the state if it is a business.
• During or after the message: a callback number, not the number of the autodialer and not a 900 number.
• For sales messages: within two seconds of the identification, an automated voice or key-press opt-out that adds you to the company’s do-not-call list and ends the call.
• If left on voicemail: a toll-free number you can call later to reach that opt-out.

A recording that starts with “Hi, this is Sarah from the benefits department” and never names a company breaks the identification rule. A caution: the statute’s own identification standard sits in section 227(d)(3), and at least one appellate court, Maryland’s Court of Special Appeals, has said section 227(d) “does not provide for a private right of action.” Missing disclosures are usually most useful as evidence alongside a consent claim under 227(b), which does carry damages.

Recorded calls that are allowed without written consent

Some prerecorded calls to home lines are exempt from the written-consent rule, but only within limits (64.1200(a)(3)):

A fourth “informational” recording in 30 days, or any recording after you opted out, falls outside the exemption. Cell phones have a separate, narrower set of free-to-end-user exemptions in 64.1200(a)(9) for package delivery, bank fraud alerts and health care reminders.

Abandoned calls and dead air

If a sales call connects and no one is there, the caller may have broken the abandoned-call rule. Telemarketers must connect you to a live person within two seconds of your greeting or play a recorded message naming the seller with an opt-out, and may abandon no more than 3 percent of answered calls in a campaign over 30 days (64.1200(a)(7)). They must also let an unanswered call ring at least 15 seconds or four rings (64.1200(a)(6)).

Ringless voicemail

Some marketers drop recorded messages straight into your voicemail without ringing the phone. The message itself is a recording, so the consent and identification rules above are the natural yardstick, although callers dispute whether a message that never rings the phone is a “call.” Our page on ringless voicemail lawsuits covers that practice.

How to prove it was a recording or AI

• Save the voicemail. It is the best evidence of a recorded message. Export it before your carrier deletes it.
• If you answer, note whether the voice paused unnaturally, repeated itself, ignored your questions, or answered off-topic.
• Write down any company name, website or callback number the message gave.
• Screenshot the call log showing the date, time and number.
• If the voice sounded like someone you know asking for money, hang up and call that person directly. Report it to the FCC and your state attorney general.

Track each call in our robocall evidence log. Each unlawful recorded call is a separate violation, so the count matters. The damages page shows how the numbers add up.

What to do about recorded or AI robocalls

If you never gave the company permission, and the call played a recording or a synthetic voice, you may have a claim for each call. Try to learn who is selling: press the option to speak to someone and ask for the company name, or note the website the recording mentions. Common sources of these calls include auto warranty robocalls and Medicare plan robocalls. Then read our guide on suing a telemarketer for the steps from here.