September 14, 2026 · 5 min
A Creator Wants to Use AI to Make Your Ad. Should You Let Them?
A creator you booked sends a note before the shoot. They want to use AI for part of the video. Maybe it is the voiceover, maybe it is a background, maybe it is the whole edit. You have a launch date, a brief they already agreed to, and no policy for this.
The answer is usually yes, with limits, and the limits have to be written down before anything is made. AI in the production process is not the problem. The problem is AI that creates an impression the audience cannot verify: a person who does not exist, a result that never happened, or a demonstration of your product that nobody actually performed. Draw the line there and most of the difficulty disappears.
What does a creator mean when they say they will use AI?
The phrase covers at least five different things, and they carry very different risk.
- Production help. Script drafts, captions, subtitles, colour grading, cutting a long take into a short. The creator still appears and still uses the product
- Synthetic voice. The creator's own voice cloned, or a generic voice reading the script over real footage
- Generated backgrounds or B roll. Real creator, invented environment or invented supporting shots
- Generated demonstration. The product shown doing something, rendered rather than filmed
- Fully synthetic creator. A character who is not a person, performing an endorsement
The first is invisible to your audience and hardly worth a conversation. The last two are where brands get into trouble. Before you answer the creator, ask which of these they mean, because most people say AI and mean the first one.
Which uses are safe to approve, and which are not?
A simple test works better than a long policy. Ask whether the AI is changing how the content was made, or changing what the content claims.
Changing how it was made is fine. Faster editing, cleaner audio, a tidier script. Nothing the viewer believes about your product depends on it.
Changing what it claims is not fine. If AI produces the before and after, the crowd in the room, the speed of the result, the texture of the food, or the person recommending it, then the viewer is being shown something that did not occur. That is a claim problem before it is an AI problem, and claim problems are the ones that reach regulators.
The middle cases are voice and appearance. A cloned voice over real footage of the real creator is generally acceptable if the audience knows the creator and the words are genuinely theirs. A generated face is not, at least not without saying so plainly on screen.
What do the disclosure rules expect?
Two disclosures now travel together in sponsored content. The paid relationship has always needed to be clear and obvious. On top of that, current guidance and recent state law expect the audience to be told when an endorser is not a real person, or when material parts of what they are seeing were generated. Dynamis LLP sets out the current position, and New York's synthetic performer disclosure law took effect in June 2026, which means a national campaign has to meet the strictest standard that applies to any viewer it reaches.
The part brands miss: a clause in your contract that makes the creator responsible does not move the liability away from you in the eyes of a regulator. It gives you a claim against the creator afterwards. It does not stop the letter arriving. So treat disclosure as something you check before the post goes live, not something you delegate and forget. If a campaign involves synthetic people or generated product demonstrations, run the plan past your own legal counsel rather than relying on a general article.
What to put in the brief and the contract
You do not need a new legal document. You need four lines in the brief you already send.
- A definition. Say what you consider AI generated for this campaign, using the five categories above rather than the word AI on its own
- A permission list. Name what is allowed without asking, what needs written approval, and what is never allowed
- A disclosure requirement. State that any allowed generated element is labelled in the post itself, in the way the platform provides, and that the paid partnership tag stays on regardless
- A delivery requirement. The creator confirms in writing, when they submit the file, which elements were generated
That fourth line is the one people skip and the one that saves you. It creates a record, and it makes the creator think about the question before delivery rather than after publication.
How do you check the finished file?
You cannot detect every generated frame, and trying is a waste of a launch week. Check the things that matter instead.
- Does anything in the video demonstrate a product outcome? If yes, ask how it was filmed and who was holding the camera
- Does the creator appear in a continuous take using the product, or only in cutaways
- Does the voice match their other recent posts in rhythm and breathing, not just in tone
- Is the platform's own AI label applied where the platform requires it
- Does the caption disclosure survive a mobile view without tapping more
If two of those raise a question, ask for the raw file. A creator working honestly will send it.
When should the answer be no?
Say no when the product is one where a viewer's trust carries real consequence: anything ingested, anything medical, anything financial, anything sold to children. Say no when the generated element is the proof, because then the ad is a claim with nothing behind it. And say no when you cannot get a straight answer about what was generated, which is rarer than you expect and tells you everything.
The Bottom Line
Let creators use AI to make content faster, not to make claims you cannot back. Define the categories in the brief, require written approval for voice, faces and demonstrations, require disclosure in the post itself, and ask the creator to declare what was generated at delivery. Remember that the disclosure duty follows your brand even when the creator publishes, so review before launch rather than trusting a contract clause to protect you afterwards.
The simplest protection is still working with creators whose past branded work you have actually checked, which is the part ViralFusion does before anyone is recommended.