TCPA

Trust, Privacy & ComplianceTCPAalso: Telephone Consumer Protection Actalso: prior express written consent

Not legal advice. This page describes legal and regulatory frameworks in general terms. Jurisdiction changes the answer, and the same voice deployment can be lawful in one market and unlawful in another. Take advice on your specific circumstances.

In one sentence

The TCPA is the United States law restricting calls and texts made with autodialers or with an artificial or prerecorded voice, and since a February 2024 FCC declaratory ruling it plainly covers AI-generated voices.

Prior express written consent Telephone Consumer Protection Act, 2026Last reviewed 30 July 2026

Not to be confused with Consent and call recording law, STIR/SHAKEN and spam-likely labeling, or AI disclosure and the EU AI Act.

Definition

The TCPA is the US law about calling and texting people who did not ask to be contacted.

It matters more to AI voice than any other rule in this category. A federal regulator has now placed AI-generated voices squarely inside the restriction on artificial and prerecorded voices.

The Telephone Consumer Protection Act restricts calls placed using automatic telephone dialing systems and calls using an artificial or prerecorded voice. The strictest requirements land on marketing calls to mobile numbers.

The core requirements

  • For marketing calls and texts to mobile numbers placed with an autodialer or an artificial or prerecorded voice, prior express written consent is generally the requirement.
  • Prior express consent, the lower standard, is generally enough for non-marketing informational calls.
  • Do-not-call obligations, covering both the national registry and the internal list.
  • Identification and opt-out requirements during the call.
  • Calling time restrictions.

The ruling that changed AI voice exposure

  • In February 2024 the FCC issued a declaratory ruling confirming that AI-generated voices in calls fall within the restriction on artificial or prerecorded voices. Any ambiguity about whether a synthesized conversational voice was covered is resolved. It is.
  • The practical effect is that an outbound AI voice agent making marketing calls to mobile numbers without prior express written consent is exposed, however natural or interactive the conversation sounds.

Why the liability profile is unusual

  • The statute provides a private right of action with statutory damages, commonly cited as USD 500 per violation and up to USD 1,500 for willful or knowing violations.
  • Per-violation damages aggregate rapidly across a calling campaign, and class action exposure is a well-established feature of this area.
  • It is one of the few compliance areas where the arithmetic on its own can be existential for a mid-sized business.

What it does not reach

  • Inbound calls. Somebody calling you is not a TCPA problem.
  • On-site voice interactions. A website widget makes no call, dials no number, and is not within scope.
  • This is a substantial structural advantage of inbound and on-site voice over outbound calling, and it is rarely spelled out.

Consent management is the operational requirement

  • Written consent has to be obtained, recorded, retained and honored on revocation.
  • Revocation must be honored promptly and through any reasonable method.
  • Consent records are the evidence in litigation, which makes consent logging an engineering requirement rather than a marketing one.

Common misconception

That giving a voice agent a natural, conversational manner lifts it out of the artificial-voice restriction. The FCC ruling runs the opposite way. How good the synthesis sounds is beside the point. That it is synthesized at all is the point.

Why it matters commercially

For any US outbound voice deployment the TCPA is the dominant legal risk, and it belongs in the assessment before the build rather than after it. It is also the strongest available argument for inbound and on-site voice, which sit entirely outside it.

In voice specifically

This is the rare case where better speech technology increases legal exposure rather than reducing it. A synthesized voice good enough that the recipient believes they are speaking to a person is still a synthesized voice, and the statute keys on the synthesis rather than on how convincing it is. Every improvement in naturalness leaves the exposure exactly where it was.

Where AsqVox fits

The TCPA does not reach the Orb. A voice widget running in the browser places no call and dials no number, so the framework has nothing to attach to. It is worth stating plainly in any comparison with an outbound voice agent vendor, because a buyer researching the category will meet TCPA warnings before they meet this page.

Visual

What the TCPA reaches, and what it does not

What the TCPA reaches, and what it does notInside TCPA scopeA call is placed and a number is dialedOutbound marketing call to a mobile: prior express written consent requiredOutbound informational call: prior express consentAI-generated voice on an outbound call: FCC declaratory ruling, February 2024, confirmed AI voices are coveredText message campaignsOutside TCPA scopeNo call placed, no number dialedInbound call: the customer called youWebsite voice widget: no call placed, no number dialedWhat decides which side you are onWho started the contactWhether a number was dialedWhether the purpose is marketingWhether the recipient is a mobile numberDamages, which is why the arithmetic is the riskUSD 500 per violationUp to USD 1,500 if willful or knowingMultiplied across the volume of a campaignPrivate right of action, with well-established class action exposure

The strongest argument for inbound voice is that this page does not apply to it.

Sounding natural does not help. The test is that the voice is synthesized, not how good the synthesis is. The two columns are drawn the same size, and what separates them is who started the conversation.

Statistics

Every figure carries its source and year. Vendor numbers are labelled as vendor numbers, and where no reliable figure exists this page says so rather than borrowing one.

The TCPA restricts calls using automatic telephone dialing systems and calls using an artificial or prerecorded voice, with prior express written consent generally required for marketing calls to mobile numbers.

Prior express written consentindustry range

Telephone Consumer Protection Act, 2026 - A statutory fact rather than a measurement. The lower prior express consent standard generally applies to non-marketing informational calls, and the difference between the two is where most consent audits fail.

In February 2024 the FCC issued a declaratory ruling confirming that AI-generated voices fall within the artificial or prerecorded voice restriction.

February 2024industry range

US Federal Communications Commission declaratory ruling, 2024 - Cite the date. It is the single fact that turned AI voice from an open question into a settled one, and it is recent enough that litigation applying it is still accumulating.

The statute provides a private right of action with statutory damages commonly cited as USD 500 per violation, rising to up to USD 1,500 for willful or knowing violations.

USD 500 to 1,500 per violationindustry range

Telephone Consumer Protection Act, 2026 - Per violation, not per campaign. Across a calling campaign the multiplication is the exposure, and class action exposure is a well-established feature of this area.

STIR/SHAKEN attestation and spam labeling separately determine whether outbound calls are answered at all. Compliance does not guarantee deliverability.

Two independent gatesindustry range

STIR/SHAKEN framework and carrier analytics, 2026 - Adjacent context rather than a TCPA fact. An outbound campaign has to clear a legal gate and a network gate, and clearing one says nothing about the other.

TCPA case law and regulatory interpretation are active and evolving, including on the definition of an autodialer and on the mechanics of revocation.

-no reliable figure

Link to current regulatory sources rather than treating any summary, this one included, as settled. Litigation continues to shape it.

No reliable aggregate statistic exists for TCPA actions specifically involving AI voice agents, because the FCC ruling is recent.

-no reliable figure

The absence is a function of timing rather than of safety. A thin case record is not evidence of low risk, and the damages arithmetic is unchanged either way.

Examples

In practice

A company builds an outbound AI voice agent for lead follow-up, reasoning that a genuinely interactive conversation is not a prerecorded message and the artificial-voice restriction therefore does not bite. Legal review points to the FCC declaratory ruling of February 2024 and pauses the campaign for a consent audit. The consent gathered through the web form falls short of the prior express written consent standard that marketing calls to mobiles demand. The build was never the problem. The consent architecture was.

The everyday version

The TCPA is the American law covering calls to people who never asked to hear from you. The regulator settled in 2024 that an AI voice is caught by it, so a human-sounding voice buys nothing. Because the penalty is counted on every single call, a marketing campaign is the rare place where the legal bill alone can get frightening fast. None of it bites when the customer arrives at your website and opens the conversation.

Usage

Who says it

  • US legal counsel and compliance teams, who follow the evolving case law closely.
  • Outbound campaign operators and contact center compliance, routinely.
  • Voice AI vendors selling outbound capability into the US, who have to be fluent in it.

Where it turns up

  • Usually in the same clause as consent capture and storage, do-not-call list management, revocation handling, calling time controls, and call identification and disclosure.
  • In US outbound procurement it is a primary diligence area, often examined more closely than the AI capability itself.

Common misuse

  • Assuming a genuinely interactive AI conversation escapes the artificial-voice restriction. The FCC ruling speaks to that directly.
  • Taking a web form's consent to be prior express written consent for marketing calls without checking what that standard actually requires.
  • Raising TCPA worries about inbound or on-site voice, which sit outside it, or missing the chance to flag that distinction when it is a real advantage.

Questions people ask

Does the TCPA apply to AI voice agents?

Yes, for outbound calls. In February 2024 the FCC issued a declaratory ruling confirming that AI-generated voices fall within the restriction on artificial or prerecorded voices. An outbound AI voice agent making marketing calls to mobile numbers without prior express written consent is exposed, regardless of how natural or interactive the conversation is.

Does the TCPA apply to a website voice widget?

No. The statute reaches calls that are placed and numbers that are dialed, and a widget running in the browser does neither. An inbound call is not a TCPA problem either, because the customer initiated it. This is a structural advantage of inbound and on-site voice over outbound calling, and it is rarely stated.

What are TCPA damages per violation?

Statutory damages are commonly cited as USD 500 per violation, rising to up to USD 1,500 for willful or knowing violations, and the statute provides a private right of action. Damages are counted per call, so exposure across a campaign multiplies rapidly, and class actions are a well-established feature of this area.

Does making an AI voice sound human help with the TCPA?

No, it points the wrong way. The FCC ruling keys on the fact that the voice is synthesized, not on how good the synthesis is. A conversation convincing enough that the recipient believes they are talking to a person is still an artificial voice under the statute, and consent obligations are unchanged.

Share this definition

Last reviewed 30 July 2026. Written and reviewed by Dhruv Dholakia, founder of AsqVox.