“AI-generated voice” is not a permission slip. An ad can carry that label and still mislead listeners about who endorsed the product, whether the testimonial is real, or what the product can do.
Synthetic voice advertising needs separate checks for disclosure, consent, and truthfulness. The policies below are named and dated where the publisher provides a date. They are a production checklist, not legal advice or a guarantee of approval. Rules change, and you must check the current requirements of every destination before launch.
Which disclosure question should you answer first?
Start with the listener's likely impression: who appears to be speaking, and what does the voice imply? A neutral synthetic announcer describing an offer is different from a clone of a recognizable person saying they personally use it.
Then identify the campaign type. Ordinary commercial advertising, political advertising, organic branded content, and broadcast spots can follow different requirements. Approval in an organic upload workflow doesn't necessarily satisfy the paid advertising workflow.
Keep sponsorship disclosure separate from AI disclosure. The former explains the commercial relationship; the latter explains how the media was created or altered. You may need both. Neither establishes permission to use a person's identity.
For a practical campaign workflow, our advertising production page covers creating the audio. The policy review belongs before generation, while you can still change the premise rather than patching a finished spot with hurried disclaimers.
What do Google, Meta, and TikTok currently require?
Don't treat the platforms as a single rulebook. Their labels, detection methods, and advertiser controls differ, and a general AI label can coexist with stricter political-ad rules.
| Destination | Dated source | Production implication |
|---|---|---|
| Google advertising | TechCrunch, July 9, 2026, reporting Google's new disclosure controls | Check the AI creation control and the consumer-facing ad information, including when creative was generated outside Google |
| Facebook and Instagram ads | Social Media Today, July 7, 2026, reporting Meta's updated AI info labels | Automatic labels can depend on the creation tool or detected metadata; don't assume an audio export will label itself |
| TikTok ads | TikTok Misleading and False Content policy, updated April 2026 | Completely generated audio and voice-cloned speech fall within the significant-edit disclosure examples |
Sarah Perez's July 9, 2026 TechCrunch report describes Google's “How this ad was made” information in My Ad Center. It reports automatic disclosure for Google's own generative advertising tools and an advertiser control for externally generated creative. Check the live Google Ads interface and current policy rather than assuming local generation is automatically detected.
Andrew Hutchinson's July 7, 2026 Social Media Today report quotes Meta's use of automatic AI info labels for certain generative tools and detected third-party metadata. The report focuses on image and video edits. It does not establish that every synthetic voice track receives an automatic label or that every audio-only use has an identical disclosure requirement.
TikTok's April 2026 policy is explicit: significantly edited or generated content requires the AIGC label or a clear disclaimer, caption, watermark, or sticker. Its examples include fully generated audio and making someone say something they didn't say through voice cloning. Undisclosed content can be rejected or restricted.
For political or issue advertising, stop and consult the destination's dedicated rules, geographic restrictions, and verification process. The commercial checklist above is not political-ad clearance.
What do broadcast and advertising bodies expect?
In the UK, CAP and BCAP govern non-broadcast and broadcast advertising respectively. CAP's Disclosure of AI in Advertising guidance, dated May 29, 2025, says the codes did not contain AI-specific rules and that existing rules apply regardless of how an ad is generated.
That dated guidance rejects a simplistic label-everything approach. It asks whether omitting the AI information would mislead the audience and whether disclosure clarifies the message or contradicts it. It also warns that disclosure is unlikely to cure a fundamentally misleading claim. Confirm the current position with your broadcaster or clearance contact before transmission.
In the US, the FTC's Endorsement Guides, revised effective July 26, 2023, address honest endorsements, substantiated claims, and clear disclosure of relevant material connections. They are not a universal AI-voice labeling law. Their relevance is what the ad leads consumers to believe.
These sources are not worldwide broadcast approval. A station, network, jurisdiction, or regulated product category may impose additional requirements. Give the broadcaster the synthetic-voice details and ask for written clearance rather than assuming platform acceptance covers radio or television.
What should a brand-safety reviewer look for?
Our recommended review starts with meaning, not audio quality. Could a reasonable listener mistake the voice for a real customer, expert, celebrity, or employee? Does the script imply firsthand experience that nobody can verify?
Ask for evidence behind product claims. A warm voice can make an unsupported claim sound like friendly advice, but that doesn't make it less consequential. Check the landing page against the spoken offer, including restrictions and eligibility. TikTok's April 2026 policy expressly prohibits inconsistent promotional information between ads and their destinations.
Also review the context surrounding the voice. A news-style sound bed, an official-sounding introduction, or an invented professional title can create an impression the script never states directly. Don't limit review to literal sentences.
Then listen for the disclosure itself. It must survive ordinary playback, background music, and the final edit. A line that exists in the script but disappears beneath the mix is not a useful safeguard. Record the review decision and the exact asset approved.
How do you document consent for a cloned advertising voice?
A recording file is not consent. Neither is a general agreement to appear in content. Obtain specific authorization to create synthetic speech and use it in the planned advertisements.
Document the speaker's identity, authority to grant permission, permitted brands, product categories, territories, destinations, and usage period. Include whether translations, new scripts, and future campaigns are allowed. Define compensation, approval rights, access to the clone, and what happens when the relationship ends.
Keep the signed agreement with the source recording's provenance and the approved scripts. Establish a contact for withdrawal or disputes and a process for stopping future use. A lawyer should adapt the agreement to the relevant jurisdiction and employment or talent arrangements.
Read the Vois voice-cloning terms before using the feature. Our voice-cloning guide explains the production side; neither replaces permission from the person whose voice you intend to use.
How do you keep approvals attached to the final ad?
Create a release record containing the script revision, voice authorization, destination, claim evidence, disclosure wording, reviewer, and final filename. Keep the uploaded version linked to that record, not merely a similarly named preview.
Local voice cloning changes where generation happens. It does not create a right to impersonate someone or remove platform obligations. If the brand changes the claim, destination, or speaker treatment, reopen approval before regenerating the passage.
Check the destination controls again at upload. Preview the finished ad with its label and landing page, then keep evidence of the settings used. Assign responsibility for campaign expiry and policy rechecks instead of leaving them to whoever last edited the audio.
Make the voice persuasive. Make its permission and purpose unmistakable.
The Vois Team