The FCC's AI-Voice Ruling, Explained for Agents (Not Lawyers)

If an AI can make your follow-up calls, the first question isn't whether it sounds natural — it's whether the call is legal. In February 2024 the FCC answered the core question, and the answer was short: yes, the robocall rules apply. Here's what the ruling actually says, what it means before you let any AI dial a lead, and what has changed since — in plain English, for working agents.

What the FCC actually ruled

On February 2, 2024 (released February 8, 2024), the FCC adopted a declaratory ruling in CG Docket No. 23-362 — the document is FCC 24-17 — holding that AI-generated voices are "artificial voices" under the Telephone Consumer Protection Act (TCPA).

Note the word declaratory. The FCC didn't write a new law; it clarified that a law from 1991 already covers voice-cloning and AI-generated speech. Every rule that has always applied to prerecorded and artificial-voice calls — consent requirements, identification requirements, penalties — applies to AI voices, and applied before the ruling too. The ruling closed the "but it's AI, not a recording" loophole before it opened.

The consent rules, translated

For an agent, the practical core is consent, and the bar depends on what the call is:

• Calls to cell phones using an artificial or AI voice require prior express consent — the person agreed to be called.
• Marketing calls are held to a higher bar: prior express written consent. If the AI call promotes your services — and prospecting calls generally do — you need a signed, written agreement (electronic counts under E-SIGN) in which the person expressly agreed to receive calls made with an artificial or prerecorded voice from you, at that number.

And the exception agents most want to exist, doesn't:

• An existing business relationship does not exempt you. The FCC eliminated the established-business-relationship exemption for prerecorded telemarketing calls in a 2012 order, effective October 2013. "They're a past client" or "they inquired once" is not written consent.

The good news: for online leads, written consent is usually solvable at the point of capture. A properly worded, clearly disclosed consent checkbox on the form where the lead gives you their number can satisfy the written-consent requirement. A phone number scraped from a sign-in sheet cannot.

What it costs to get this wrong

The TCPA has teeth of a specific kind: a private right of action. Recipients don't have to persuade a regulator to care — they can sue you directly, and there's an established plaintiffs' bar that does little else.

• $500 per violation — per call, not per campaign.
• Up to $1,500 per willful or knowing violation.

Multiply by a contact list and the arithmetic explains why compliance-by-design matters more than call volume.

What has changed since the ruling

The 2024 ruling was a beginning, not an ending. Three developments matter as of mid-2026:

• The one-to-one consent rule was struck down. The FCC had adopted a rule requiring lead-generation consent to name one seller at a time (no more "and our 400 marketing partners"). The Eleventh Circuit vacated it on January 24, 2025, and the FCC formally removed it from the rules on August 29, 2025. The baseline written-consent requirement still stands — what fell was the stricter one-seller-per-consent add-on.
• An AI-disclosure rule is still pending. The FCC opened a rulemaking (FCC 24-84) proposing that AI-generated calls explicitly disclose they're AI at the outset. As of mid-2026 it remains a proposal, not a rule — but it tells you exactly where federal policy is heading.
• States are moving faster. California's AB 2905, effective January 1, 2025, requires automated announcement calls to disclose when the voice is AI-generated. Other states have their own disclosure and mini-TCPA laws, and they don't wait for Washington.

A practical checklist before any AI makes a call for you

1. Call your own leads, not lists. People who contacted you and consented — never purchased or scraped numbers.
2. Capture written consent where the lead is captured. Clear language, checkbox, timestamped record.
3. Have the AI identify itself. Federal law already requires identifying who's calling; a pending federal rule and live state laws point at explicit AI disclosure. Building it in now is cheap; retrofitting under a lawsuit is not.
4. Honor opt-outs and do-not-call requests immediately, and scrub against the DNC registry for anything prospecting-shaped.
5. Keep records — of consent, of calls, of disclosures. In a TCPA dispute, the paper trail is the defense.

Where Marshal stands on this

We built Marshal's AI calling assuming the strictest reading would eventually win. Marshal calls only your own leads — people already in your book, with their consent status in view — the AI voice discloses what it is at the start of the call, and every call is logged on the client's timeline. When a call goes well, it books the appointment straight onto your Google Calendar. The point of an AI assistant is to take work off your desk, not to put legal risk on it.

The full legal walkthrough — consent language, record-keeping, and how the rules apply call-type by call-type — lives in our guide to TCPA rules for AI calling, and the state-by-state picture (California is not alone) is in AI disclosure laws by state. If you're weighing whether AI calling belongs in your follow-up at all, start with what an AI ISA actually does.

This article is general information, not legal advice. Telemarketing law is fact-specific and changes; talk to a lawyer licensed in your state before running any calling program.