Compliance

TCPA and AI voice calls: what did the FCC actually decide?

Updated

In February 2024 the FCC declared that the TCPA’s restrictions on the use of an “artificial or prerecorded voice” encompass current AI technologies that generate a human-sounding voice. It did not prohibit those calls. It placed them inside a body of rules that has governed prerecorded robocalls since 1991 — consent before the call, identification during it, an opt-out that works, and a set of mechanics that binds whatever the recipient agreed to.

So the useful questions are narrower than “is AI calling legal.” They are: what does the ruling actually say, which rules apply even when the consent is perfect, is there a law requiring the call to announce that it is AI, and what do individual states add on top. All four are answerable from primary sources, and all four are answered below.

This is not legal advice

Everything here is a description of what a named statute, regulation or FCC order says, with a link to the source so you can read it yourself. It is not a legal opinion, it is not advice, and it does not tell you whether your own script, consent record or dialling list satisfies any of it. Those are questions for a TCPA attorney looking at your actual campaign.

What did the FCC's February 2024 AI-voice ruling actually decide?

Three things, and the third is the one people miss. The declaratory ruling in CG Docket No. 23-362, adopted on 2 February 2024 and released on 8 February, holds that the TCPA’s restrictions on the use of “artificial or prerecorded voice” encompass current AI technologies[1] that resemble human voices, and that callers must secure prior express consent before making such a call.

Second, the identification and opt-out duties travel with it. The same ruling records that “all artificial or prerecorded voice messages must provide certain identification and disclosure information for the entity responsible for initiating the call”[2], and that where a message includes or introduces an advertisement or constitutes telemarketing, the caller must offer specified opt-out methods.

8 Feb 2024
Release date of the declaratory ruling that brought AI-generated voice inside “artificial or prerecorded voice”
Source: FCC 24-17, Declaratory Ruling
$500
Statutory damages per violation under 47 U.S.C. § 227(b)(3), which a court may increase to not more than three times that amount
Source: 47 U.S.C. § 227(b)(3)
Two prongs
§ 227(b)(1)(A)(iii) restricts an autodialer or an artificial or prerecorded voice — independently of each other
Source: 47 U.S.C. § 227(b)(1)(A)(iii)

Third, and structurally: the artificial-voice prong stands on its own. Section 227(b)(1)(A)(iii) reaches a call to a cellular number made using an automatic telephone dialing system or an artificial or prerecorded voice, so the years of argument about what counts as an autodialer decide nothing here. A carefully non-automatic dialler placing AI-voice calls is inside the statute on the second prong alone.

Consent answers the question of whether a call may be placed at all. It answers nothing about how the call has to behave once it connects, and that second set of duties is where operating systems fail rather than where forms fail. What the consent itself has to contain is a separate question, covered on what your opt-in form must say.

Federal duties that attach to an artificial-voice call independently of consent
RuleWhat the text requiresWhere it lives
Identification“At the beginning of the message, state clearly the identity of the business, individual, or other entity that is responsible for initiating the call.”47 CFR § 64.1200(b)(1)
Opt-out mechanism (advertisement or telemarketing calls)For a call that includes or introduces an advertisement or constitutes telemarketing: an automated, interactive voice- and/or key press-activated opt-out mechanism, offered during the call itself. The rule does not attach to a purely informational call.47 CFR § 64.1200(b)(3)
AbandonmentA call counts as abandoned where a live representative is not connected within two seconds of the called party's greeting, with abandoned calls capped at 3% measured over 30 days.47 CFR § 64.1200(a)(7)
Calling hoursNo solicitation “before the hour of 8 a.m. or after 9 p.m. (local time at the called party's location)”.47 CFR § 64.1200(c)(1); 16 CFR § 310.4(c)
RevocationRevocation by any reasonable means, honoured “within a reasonable time not to exceed ten business days from receipt of such request”.47 CFR § 64.1200(a)(10)
Do-not-call requestsA do-not-call request must be honoured “for 5 years from the time the request is made”.47 CFR § 64.1200(d)(6)

Two of those rows carry numbers, and both belong to the regulator rather than to us. The connect-time and abandonment figures describe the rule’s own test for an abandoned call[3]; they are not a performance claim, and you will not find a timing figure of ours anywhere on this site. The calling-hours rule is worth reading twice because it is measured at the called party’s location, which is not the caller’s time zone and not reliably the area code’s either — and the FTC’s Telemarketing Sales Rule imposes the same window, plus prompt oral disclosures of the seller’s identity and that the purpose of the call is to sell[4], under a different statute entirely.

Is there a law requiring the call to say it is AI?

Not as an adopted federal rule, which surprises people who assume the 2024 ruling created one. It did not. What exists federally is a proposal: FCC 24-84, a notice of proposed rulemaking released in August 2024, would amend § 64.1200(b)(1) so that a message must “state clearly the identity of the business, individual, or other entity that is responsible for initiating the call, and disclose whether the call uses an artificial intelligence-generated voice”[5], alongside a proposed definition of an AI-generated call covering technology that produces voice or text content to communicate with a called party on an outbound call. Proposed is the operative word in both cases.

States legislate here independently, and California is the one that reaches an artificial voice by name. Public Utilities Code § 2874 provides that where calls are placed through an automatic dialing-announcing device, the device may be operated only after an unrecorded, natural voice announcement[6] that states the nature of the call and the caller’s name, address and telephone number, inquires whether the person called consents to hear the prerecorded message, and informs them if the prerecorded message uses an artificial voice.

Two scope questions decide whether § 2874 reaches a conversational agent — and we take no position on either

First, the device definition. § 2871 defines an automatic dialing-announcing device around equipment with the capability “to disseminate a prerecorded message to the telephone number called”[7], and speech generated in real time is textually not a prerecorded message — while the legislature’s later insertion of an artificial-voice disclosure into the same article points the other way. Second, § 2872 exempts a device “not used to randomly or sequentially dial telephone numbers, but that is used solely to transmit a message to an established business associate, customer, or other person having an established relationship”[8], and any call generated at the recipient’s request. Neither question is settled by any authority we located, and this page does not resolve them in either direction. It is the question to put to California counsel before a first call to a California number, not after.

Separately from all of that, our own assistant says it is an assistant on every call. That is a product decision rather than a reading of the law — buyers asked for it, and it removes exposure rather than creating any. The exact wording is on what it says. Whether disclosure changes how recipients respond is untested: we found no public body of evidence either way, and we are not going to invent one.

What do the state mini-TCPA laws add — Florida, Washington, Oklahoma?

Federal law is the floor. Three states were worked from primary statutory text for this page, and they fail in three different ways, which is the more useful finding than any single rule. Washington prohibits rather than regulates: RCW 80.36.400 defines an automatic dialing and announcing device as “a system which automatically dials telephone numbers and transmits a recorded or artificial voice message once a connection is made”[9] and says no person may use one for purposes of commercial solicitation — with commercial solicitation defined as the unsolicited initiation of a call encouraging a purchase.

Three states worked from primary sources, and one gap we are not filling
StateMechanismDoes the text name an artificial voice?
CaliforniaDisclosure plus a live, unrecorded natural-voice announcement before an automatic dialing-announcing device operates — subject to the § 2872 exemption.Yes — the announcement must inform the person if the message uses an artificial voice.
WashingtonA flat prohibition on using such a device for commercial solicitation. No consent exception appears in the text; the statute turns on whether the call was solicited.Yes — “recorded or artificial voice message” sits inside the definition itself.
FloridaA consent requirement: prior express written consent for telephonic sales calls using an automated system, with a private right of action.No AI-specific disclosure appears in § 501.059.
OklahomaNot verified. The operative statute text could not be retrieved when this was compiled, and we are not summarising a statute we have not read.Unknown, and recorded as unknown.
$500
Florida's statutory damages under the FTSA, which a court may increase to an amount equal to not more than three times the award
Source: Fla. Stat. § 501.059
15 days
Florida's cure window: after a STOP notice the solicitor must cease sending text solicitations to that party
Source: Fla. Stat. § 501.059
3 of 50
States read from primary statutory text for this page. The rest were not surveyed, and no row was written from memory

Florida is the state most often described in vendor marketing as a catastrophe, and its current text is narrower than that reputation: the definition turns on an automated system for the selection and dialing of telephone numbers[10], and the statute carries a fifteen-day cure window for texts after a stop request. That is a description of the text, not a view on whether any particular campaign clears it.

The fault line to watch is one word, and it moves in one direction

Older statutes say “prerecorded,” which gives a real-time conversational agent a textual argument. Newer and amended ones say “artificial,” or define an AI voice outright, and close it. Any position that depends on “we are not prerecorded” is a position with a shelf life measured in legislative sessions rather than a foundation, and the practical consequence is that which state a number sits in is an operating question rather than a paperwork one.

What should you ask any vendor placing AI calls for you?

Law is law and practice is practice, so this last part is questions rather than rules. They are the ones we would expect to be asked, and a vendor who cannot answer them from their own system rather than from a policy document has told you something useful.

1

Whose name is stated at the start of the call, and does it match the consent form?

The identification duty attaches to the entity responsible for initiating the call, and a consent form names a specific seller. If those two names are different, ask why before anything dials.

2

Where is the clock checked — at dial time or at queue time?

Calling hours run against local time at the called party's location. A call queued inside the window and placed outside it is still outside it, so the answer has to be dial time.

3

What counts as an opt-out, and how far does it travel?

Revocation may be made by any reasonable means, and a caller may not designate one exclusive channel for it. Ask whether an opt-out spoken mid-conversation, typed in a free-text reply, or sent by email all land in the same place — and whether the system argues back.

4

Can the consent record be produced, per number, months later?

The burden of proving consent sits with the caller. A consent that cannot be produced is functionally a consent that does not exist, which makes retention an evidence question rather than an administrative one.

5

Can a state be switched off, per client and per call type?

At least one state's statute is a prohibition rather than a consent rule, so the remedy there is exclusion rather than a better form. Whether that is a configuration change or a development project is worth knowing before it is needed.

None of that makes a campaign compliant, and nothing on this page asserts that ours is. It is a description of the artifacts the rules ask about. The consent side is covered on is it legal to call and text registrants, and the question of which party carries the exposure is on who is liable.

Frequently asked questions

If you arrived here because you received one of these calls, the two documents worth asking about are the consent record and the suppression log. If you arrived as an operator, the honest next step is a TCPA attorney and your own campaign, not a vendor page. What our assistant says word for word is on what it says. Once more, plainly: this page is not legal advice, and it does not assess anyone’s compliance posture.

JB
Justas Butkus

Founder & Operator, CallHush

Founder and operator of CallHush. The offer is one sentence: you run a webinar, and we increase your show-up rate and your post-webinar sales with an AI voice and SMS system. CallHush has no closed clients yet — the first engagement is a pilot run as a split of the client’s own registrant list, and nothing on this site is presented as a client result.

View all articles