Are AI voices legal in robocalls? The FCC rule explained
Updated November 6, 2026
The picks
Press play to hear each voice read this guide's sample line.
These picks compare my fifteen voices. I own them, and ElevenLabs may pay me when you use them.
Dexter·Approved service messages
An even read for a notice after the calling workflow has legal approval. Voice choice supplies no consent.
Derek·Formal internal audition
Firm pacing makes an operational notice easy to evaluate. It can sound impersonal on a friendly support line.
Reel·Brief service wording
Conversational delivery for a short approved message. A friendly tone doesn't change the rules that govern the call.
AI voices are subject to the TCPA’s artificial-voice calling rules. In February 2024, the FCC confirmed that AI-generated human voices fall within that category. Prior express consent generally applies where those restrictions govern, with different requirements for telemarketing and specific exemptions. It was not a blanket prohibition on every use of synthetic speech.
This guide explains that ruling and the production questions I’d settle before recording. It is general information, not advice that a particular call is lawful. Get current legal review for your campaign, recipients and jurisdictions before using an outbound calling system.
What did the February 2024 ruling actually settle?
The FCC classified voices generated with AI as artificial voices for TCPA purposes. A system does not escape those rules because it creates a sentence during the call, uses a voice clone, or sounds conversational.
That gives a production team a direct starting point. If a project uses an AI voice on an outbound call, put it through the artificial-voice review. Don’t wait until someone asks whether a listener could tell it was synthetic.
The ruling works within an existing framework. Its classification is different from deciding that your particular message, recipient, phone number and consent record satisfy every applicable requirement. The official FCC announcement also discusses prior express written consent for telemarketing. A general contact permission should never be silently relabeled as permission for a sales robocall.
Which questions belong in the legal brief?
Write the brief before asking counsel for a yes or no. Include the exact message, who initiates the call, the number categories, the recipient group and the purpose. State whether the system is making a promotional offer, delivering a service notice, requesting money or discussing a political campaign. Attach the consent language and explain how you recorded it.
Keep the message itself in the brief. An appointment reminder with an added sales offer can be a different project from the reminder your team originally reviewed. The title in the workflow builder won’t settle that question.
Ask counsel to identify the applicable consent standard, any claimed exemption, required identification and opt-out behavior, and rules for revocations. Avoid assuming that residential lines, mobile numbers and all informational messages receive identical treatment. If an exemption is the reason a workflow may proceed, document who approved that interpretation and what conditions it depends on.
How would I separate service and promotional production?
I would give them different scripts, approval records and workflow names. That makes a last-minute promotional sentence easier to notice before it changes an approved service message.
Suppose your test project is a reminder that a fictional library hold is ready. Keep its script about the hold and the approved next step. If somebody adds a membership promotion at the end, return it for review. Don’t hide the new sentence in an audio replacement request.
Use a staging destination for auditions. A folder of MP3s should not automatically trigger dialing. The person approving the recording needs time to hear the words, compare them with the approved text and check the phone-format export.
For general system design, my IVR reference describes recorded phone messages. The AI phone agent guide covers turn-taking and handoffs. Neither is authorization to place a call.
What if the message is political?
Take both the calling rules and the generation service’s policy to the review. ElevenLabs’ policy requires prior written consent from ElevenLabs for political campaigning. It also prohibits voter suppression and unauthorized political impersonation. A voice’s presence in a library supplies none of those permissions.
Don’t label a campaign message “commentary” to get it through a tool. Give the reviewer the actual use and distribution plan. The political and advocacy reference explains the site’s casting context, but the decision to generate belongs after that permission check.
I would keep a candidate’s own voice out of a synthetic audition unless the necessary authorization is established. A generic narrator reading clearly attributed approved words is a different creative choice from making an identifiable person appear to speak.
Which voices can I audition while the workflow is reviewed?
Use an internal production passage with no real recipients. My picks are limited to my own catalog, and they are about intelligibility.
Dexter is my first choice for an even service read. It measured about 190 words per minute in my clips, one of my quicker voices. Its instructional tone can drain excitement from promotional copy. That is a casting limitation, not a reason to make a service message more promotional.
Derek reads at about 180 words per minute in my clips and gives a notice a formal shape. Listen carefully to organization names and callback digits. A relaxed local business may find it too official.
Reel is a touch quicker than Dexter, roughly 195 words per minute in my clips, and suits brief conversational wording. Its casualness won’t carry a solemn announcement. You can add Dexter to My Voices for the following internal sample.
This recording is for the production team’s review. The approved message should identify who is calling and explain the purpose in plain language. Check every spoken digit against the script. Then test the available callback and stop-message routes in the staging system. Keep the approved text with the final recording so a later edit can be compared with the version that received approval.
What should happen before the first real call?
Have the responsible reviewer sign off on the calling workflow and the final wording. Confirm that changes in consent or suppression records reach the system before it dials. Then listen to the exact encoded file that system will use.
Keep the approval, script revision and audio together. If the wording changes, reopen the approval. Good audio review cannot determine whether a recipient gave the necessary consent, and a legal approval cannot catch an incorrectly spoken phone number. Both checks have a job.
Sources
Producing this for a client or an organization? See AI voices for political campaigns and advocacy groups.
Related guides
Other jobs where I use the same voices.