Can You Use Someone's Likeness in an AI Ad?

Usually not, at least not safely, unless you have specific written permission. If a reasonable viewer can identify a real person in your ad and the ad helps sell something, assume you need that person's consent until a lawyer tells you otherwise.

The rule itself is simple. Following it is harder, because AI lets a team cross the line without noticing. A founder approves a voice clone for one ad, and three weeks later the agency is using it in cutdowns with lines nobody approved. A creator prompts for "a famous actor, but slightly different." A customer's written review becomes a talking-head video of that customer. Each step feels small, but together they can put a paid campaign on the wrong side of state publicity law.

This is a production-risk framework, not legal advice. Use it to screen creative and organize paperwork, and bring in counsel for high-risk or multi-state campaigns. For music, stock, and AI tool terms, see our field guide to AI video rights and licensing.

What counts as a likeness

In the US, the right of publicity is mostly a patchwork of state laws. In general, it stops people from using someone's identity commercially without permission, but each state sets its own rules on what's protected, which exceptions apply, what remedies exist, and how long the right lasts after death. The Congressional Research Service overview is a readable map of how those state laws meet AI.

California is a useful reference point because its statute is explicit. Civil Code section 3344 covers knowingly using another person's name, voice, signature, photograph, or likeness for advertising or selling without prior consent.

In production terms, treat any of these as identity if a viewer could connect them to a specific person:

• A face, including a partial face or a generated face built from someone's photos
• A voice, including a clone or a deliberate sound-alike
• A name, signature, or social handle
• A look, outfit, or set of mannerisms that only makes sense as a reference to one person

For advertisers, the practical test is recognition. If people in the comments are asking "wait, is that...?", you have a likeness problem, however different the face looks in your source files.

Editing footage vs. creating a new performance

Teams get into trouble when they treat every AI touch as the same kind of edit. The risk changes a lot depending on what the AI actually did.

Most AI ad problems sit in the move from the second row to the third. California takes that move seriously. Labor Code section 927 makes certain digital-replica clauses unenforceable when the replica stands in for work the person would otherwise have done in person, the intended uses aren't reasonably specific, and the person had no legal counsel or union representation.

Owning the footage is a separate question. You can hold full copyright in a shoot and still have no right to build a voice model from it. Copyright covers the recording, while publicity rights cover the person in it. We go through the ownership side in who owns AI-generated ads.

Green, yellow, red: screening proposed creative

These ratings are about likeness rights only. Green means lower likeness-rights risk. It isn't a legal guarantee, and it doesn't settle disclosure, which you check separately. Yellow means you need more consent or review before production starts. Red means don't launch without explicit authorization and legal review.

If you want a fully synthetic presenter, that route has its own tradeoffs. We cover them in AI spokesperson videos.

Two separate gates: permission and truth

Clearing rights tells you whether you're allowed to use the person. It tells you nothing about whether the ad is honest. You have to pass both.

The first gate is consent, which is what most of this article covers.

The second gate is truthfulness and disclosure. The FTC's endorsement guidance says endorsements must be truthful and not misleading, endorsers shouldn't describe experiences they didn't have, and material connections must be disclosed. A fully licensed face can still produce a deceptive ad. Say a paid actor's digital replica says "this cleared my skin in two weeks" and the actor never used the product. The license doesn't fix that.

Disclosure rules are their own requirement too. New York's General Business Law section 396-b currently requires a conspicuous disclosure when a business knowingly uses a synthetic performer, as the statute defines one, in an ad. Its exceptions include audio-only ads and AI used only to translate a human performer's language. Ad platforms add their own AI labeling rules on top of that. A label tells viewers what they're watching, but it gives you no permission to use anyone. Having permission doesn't excuse a missing label either. Our synthetic media primer covers the labeling side in more depth.

Weak release vs. stronger release

Here's a paraphrase of a clause that shows up in many older talent and customer releases:

"I grant the company the right to use my name, image, likeness, and voice in all media now known or later developed, worldwide, forever."

It looks airtight, but for AI work it's weak. It never says a replica can be built, and it says nothing about what that replica may say or sell. The person who signed it was picturing a camera, not a voice model reading a script they've never seen. In California, vague language like this may not hold up for new digital performances under the Labor Code provision above. Even where it would hold up, a talent who finds their clone in an ad they never saw becomes a public relations problem, and you'll handle it in public.

Specific consent answers the questions the broad clause skips:

A lawyer should draft the actual agreement. You can use this table to check whether their draft answers every question.

Famous, deceased, employee, or only a resemblance

Famous people

Famous people carry the highest exposure. They notice, they have representation, and their presence implies an endorsement. Commercial ads sit at the risky end of the spectrum. Exceptions for news, commentary, and parody don't automatically cover a product ad, so don't count on "it's a parody" to protect you.

Deceased people

Several states protect publicity rights after death, and how long that protection lasts varies. Estates license these rights and enforce them. Treat a deceased person the same way you'd treat a living one and get permission from whoever controls the rights.

Employees

Employment doesn't give you a model release. Get separate written consent for ad use, say what happens when the person leaves, and ask again before building anything from old company footage. Appearing in a product video years ago may not mean someone agreed to front your AI ads indefinitely.

Lookalikes

If your team keeps saying "it kind of looks like...", change the character. The whole value of a lookalike is that viewers recognize the real person, and that recognition is exactly what creates the risk.

The likeness clearance packet

Ask your agency or creator for this before launch, the same way you'd ask for project files. The brand keeps a copy, and the media buyer doesn't upload without it.

• [ ] A list of every real, identifiable person in the ad by face, voice, or name, or a signed statement that no real person was used or referenced
• [ ] Signed consent for each person, explicitly covering replica creation where one was made
• [ ] A scope sheet listing product, campaign, channels, paid or organic, territory, and start and end dates
• [ ] The final script with every line marked as recorded or synthetic, plus the person's sign-off on each synthetic line
• [ ] Claim substantiation, and confirmation that any first-person experience in the ad actually happened
• [ ] A source file log showing where each piece of footage and each voice sample came from, with its license
• [ ] A tool log listing which AI tools generated or altered the person, and confirmation that each tool's terms allow paid ad use
• [ ] Prompts and reference images for any generated character
• [ ] A disclosure plan covering on-screen labels required by law or platform policy, plus material connection disclosures
• [ ] Model handling details: where the voice or face model lives, who can access it, and when it gets deleted
• [ ] Version IDs for the final cuts, so the media buyer runs exactly what was cleared
• [ ] A named contact for takedown requests

An authorized agency or creator can obtain releases or license rights for you. What your production contract with them can't do is cure missing consent from the person in the ad. If that consent was never given, no clause between you and a vendor fills the gap. Your contract should say who obtains releases, who checks claims, who adds disclosures, and who keeps the records. Most failures happen in the handoffs between the brand, the agency, the creator, and the media buyer. The agency builds new cutdowns with lines the talent never approved. The media buyer keeps a winning ad running past the consent end date. A creator uses a platform avatar whose license excludes your category. Or the brand assumes the agency handled releases while the agency assumes the brand's old release covered it.

If you commission AI video ads through a marketplace like Viralix, where vetted creators deliver campaign-ready assets, put this packet on the deliverables list. The creator can document what they built, but the rights decisions still belong to you and your counsel.

If the ad is already live

1. Pause every placement, including cutdowns, retargeting, and whitelisted creator posts.
2. Preserve everything: contracts, prompts, source files, versions, approvals, and messages. Don't delete files to make the problem disappear, because that usually makes it worse.
3. Map the footprint. List every platform, ad ID, territory, agency, freelancer, and tool that touched the asset.
4. Talk to counsel before contacting the person or rightsholder, then reach out with counsel involved.
5. Replace the asset with a cleared version.
6. Document the takedown with dates, screenshots, and platform confirmations.
7. Find the handoff that let it through and close that gap.

Before you launch

1. Don't put a real, identifiable person in an ad that sells something without written consent that names AI replication, the product, and the term.
2. Any line a real person "says" synthetically needs their sign-off on that exact line, and a check that it matches what they actually experienced.
3. No clearance packet, no launch. The media buyer gets the packet before they get the upload link.