Is It Legal to Have an AI Call Your Leads?

It's the question every business owner asks us within about four minutes of hearing what we do, and they usually ask it a little sheepishly, as though it might be rude. It isn't rude. It's the right question, and a lot of people selling AI calling would rather you didn't ask it.

The short answer: yes, it's legal — and it is squarely regulated. AI voice calls are not a loophole around telemarketing law. In February 2024 the FCC closed the door on that idea explicitly, and if a vendor tells you otherwise, that's the moment to end the call.

What the FCC decided in 2024

On February 8, 2024, the FCC issued a unanimous Declaratory Ruling holding that calls using AI-generated voices are "artificial" voices under the Telephone Consumer Protection Act — the law that has governed robocalls since 1991.

That single word does all the work. The TCPA already had a well-developed set of rules for artificial and prerecorded voice calls. By classifying AI voices as artificial, the FCC didn't write new law for AI; it dropped AI into a body of law that already existed, with decades of case law behind it.

The ruling was also written to be hard to route around. It covers technology that generates a human-sounding voice regardless of how sophisticated it is — voice cloning, real-time conversational systems, large-language-model agents. The Commission was explicit that the statute doesn't allow a carve-out for technologies that claim to be the equivalent of a live agent.

"It sounds like a person, so the rules for people apply" is the opposite of how this works.

Consent is the whole ballgame

Nearly every AI calling problem you'll read about traces back to one thing: calling somebody who never agreed to be called.

Broadly, the standard turns on what the call is for:

"Prior express written consent" is a term of art, not a vibe. It means the person agreed, in a written record you can produce later, to receive calls of that kind at that number, after being told clearly what they were agreeing to. A checkbox buried in a privacy policy is not it. A purchased list is emphatically not it.

And the burden runs the wrong way from most people's instincts: if it's ever disputed, you're the one who has to show the consent existed. Statutory damages under the TCPA run from $500 to $1,500 per call, and they stack per call with no overall cap. A modest list dialed without consent is not a modest problem.

The rules move — a live example

Here's why we keep telling people to talk to a lawyer rather than trusting a blog post, including this one.

In December 2023 the FCC adopted what became known as the "one-to-one consent" rule. It would have required consent to be given to one specific seller at a time, effectively ending the practice of a single form generating consent for a stack of "marketing partners." It was set to take effect January 27, 2025.

On January 24, 2025 — one business day before it landed — the Eleventh Circuit vacated it in Insurance Marketing Coalition Ltd. v. FCC, holding that the Commission had exceeded its statutory authority because the new restrictions conflicted with the ordinary meaning of "prior express consent." The rule never took effect, and the prior express written consent standard that existed before it continues to apply.

A whole industry spent 2024 preparing for a rule that evaporated in a single afternoon. That's the tempo here. Anything you read about telecom compliance — this article included — carries a shelf life.

Disclosure, opt-outs, and the DNC list

Consent gets you the right to dial. It doesn't end your obligations.

Identification. Calls have to identify who is responsible for them. The person picking up needs to know what business is on the other end.

Opt-out. People must be able to tell you to stop, and that has to actually work — promptly, and across the system, not just for the one campaign.

Do Not Call. The federal registry still exists and still applies. So do state rules: a number of states have their own "mini-TCPA" statutes that add requirements on top of the federal ones, and some are stricter. Where your leads live matters.

On disclosure specifically, there's a legal floor and there's a decision above it. We think an AI should say it's an AI, promptly and in plain language, without being asked. Not because a rule forces that exact phrasing, but because the alternative is building a business on a person not realizing what they're talking to — and because everyone finds out eventually.

How we actually run it

Concretely, for every client we run:

• We only dial leads who gave written consent. These are people who filled out a form in a campaign we're running, agreeing to be contacted. We don't buy lists and we don't dial scraped numbers.
• The agent discloses that it's an AI. Up front, and again any time someone asks.
• It hands off to a person on request. If someone wants a human, they get one — same window, same day. Nobody has to argue with software to reach you.
• Opt-outs and Do Not Call requests are honored. Somebody says stop, it stops.
• Calling hours are 8am–8pm local, seven days. Local to the lead, not to us.
• Every call is logged to your CRM with a summary and outcome, so there's a record of what was said and what was agreed.

None of that is a differentiator we invented. Most of it is the floor. We're writing it down because "we're compliant" is easy to say and specifics are harder, and you should be asking for specifics.

Red flags when you're shopping

If you're evaluating anyone who does AI calling, the following should stop the conversation:

• They offer to supply the leads. Ask exactly where the numbers come from and what consent record exists for each one. Vagueness here is the single biggest tell.
• They pitch the AI's ability to pass as human as a feature. "Nobody can tell" is being sold to you as sophistication. It's being sold to the FCC as intent.
• They can't describe the opt-out mechanism. If they can't tell you what happens operationally when someone says "take me off this list," it probably isn't wired to anything.
• They tell you the TCPA doesn't apply to AI. It does. Since February 2024, unambiguously.
• They won't put compliance terms in the contract. Who carries the liability if a call goes out to someone who never consented? Get the answer in writing.

The reason we wrote this is not that compliance is a great marketing hook. It's that the fastest way for this whole category to get ruined is a wave of businesses dialing purchased lists with synthetic voices and generating the kind of headlines that produce new legislation. We'd rather compete against people who are doing it properly.

Sources

• Federal Communications Commission, Declaratory Ruling FCC 24-17, adopted February 8, 2024 (AI-generated voices under the TCPA).
• Insurance Marketing Coalition Ltd. v. FCC, U.S. Court of Appeals for the Eleventh Circuit, decided January 24, 2025 (vacating the one-to-one consent rule).
• Telephone Consumer Protection Act, 47 U.S.C. § 227, and implementing rules at 47 C.F.R. § 64.1200.