Often yes, if it looks real. TikTok and YouTube require a label on realistic AI-made or heavily altered video. New York requires you to disclose AI-made people in ads, with fines of $1,000 and then $5,000. Using AI to write a script or make captions generally needs no label. And an AI person posing as a happy customer isn’t a labeling question at all: don’t do it.
The test most rules come down to
Could someone believe this really happened? That’s the line TikTok and YouTube both draw. How much AI you used matters less than whether a viewer could be misled. A cartoon made entirely by AI can be fine with no label. One realistic generated shot of your shop may not be.
What each one says
| Who | The rule | If you don’t |
|---|---|---|
| TikTok | Label AI-made or significantly edited content showing realistic-looking scenes or people | It may be removed, restricted or labeled for you |
| YouTube | Disclose at upload if it makes a real person seem to say or do something they didn’t, alters footage of a real event or place, shows a realistic scene that never happened, or creates music that’s the video’s main focus | If you keep not disclosing, YouTube may add a label itself, remove the video or suspend you from its Partner Program |
| Meta | Adds “AI info” when it detects industry-standard signals or when you disclose. For ads, it labels work from its own AI tools and detects third-party ones. We found stricter rules for political and social issue ads, and no blanket duty on ordinary ads | A label may appear anyway, since many AI tools leave those signals in the file |
| X | Its posting API has a “made with AI” setting. We couldn’t read its wider rules | Unknown |
| We found no AI label field in its posting API | Unknown | |
| New York | Anyone creating an ad for a commercial purpose must conspicuously disclose an AI-made performer, if they actually know it’s there | $1,000 for a first violation, $5,000 after that |
| The FTC | No blanket ban on AI avatars, but fake testimonials are banned, whoever or whatever delivers them. It’s also illegal to write or knowingly buy fake reviews | Civil penalties for knowing violations, up to $53,088 each, per an FTC release |
Twelve everyday cases
This is our reading of those rules, and YouTube’s examples are the most detailed. It isn’t legal advice, and where a platform hasn’t spelled something out, we say so.
| What you did | Label it? | Why |
|---|---|---|
| AI wrote the script, and you filmed it all | No | YouTube lists scripts and outlines as production help |
| AI made the captions | No | YouTube lists this too |
| Color fix, beauty filter, background blur | No | YouTube calls these minor edits, as long as they don’t mislead |
| AI cleaned up your audio or sharpened your footage | No | Also on YouTube’s list |
| Your own voice, cloned, reading your own voiceover | No on YouTube | YouTube names this. TikTok doesn’t say, so check before relying on it there |
| An obvious cartoon or fantasy scene | No | Neither rule covers what no one would take for real |
| Motion graphics made by code, with your real screens | No | There’s no AI in it. Labeling it would be wrong in the other direction |
| AI-made footage of a real place, for a promo | Yes | YouTube’s own example is a generated surfer in Maui in a travel promo |
| Your real product, placed in a generated realistic room | Yes | A realistic scene that didn’t happen. Also make sure it doesn’t change what the product looks like or does |
| AI-made music as the main thing in the video | Yes on YouTube | It’s named in the rule. It’s unclear for music quietly in the background |
| An AI person presenting your product as an obvious spokesperson | Yes | Platforms treat it as realistic, and New York’s rule covers it. The FTC doesn’t ban it outright |
| An AI person saying they bought it and love it | Don’t make this | That’s a fake testimonial. A label doesn’t fix it |
New York’s rule, in plain words
It was signed on 11 Dec 2025 and took effect 180 days later, which we work out as 9 June 2026. It covers a digital creation meant to look like a human performer who isn’t an identifiable real person. It doesn’t cover ads with only audio, or AI used only to translate a real performer’s words. It also doesn’t apply to the media that carries an ad, such as a billboard company or a TV network, or to promos for films and shows that use the performer the same way. It doesn’t define “conspicuously”, so we’d put the words on screen, large, for as long as the person is there. It also doesn’t say in the text we read how it treats a business outside New York whose ad is seen there. That’s a good question for a lawyer if you use AI people at all.
Will a label hurt the ad?
It might. In one academic test of display banners, reported in the press in 2025, saying an ad was AI-made cut clicks by about 31.5%. That’s banners and not video, and we read a report of it, not the paper. It’s a reason to use AI only where it earns its place. It isn’t a reason to leave the label off, since the same test suggests that people care, and the platforms may add it anyway.
What to do
- Keep a note of what in each video was AI-made. You’ll need it for the question at upload.
- Use the setting the platform gives you: TikTok’s AI-generated content switch and YouTube Studio’s AI use setting. If an app posts for you, ask it how it sets them. TikTok’s and Meta’s posting APIs both have one.
- Say it where people can see, if there’s an AI person in the ad. A setting alone may not meet New York’s “conspicuously”.
- Never have an AI person, or a real one, say something a real customer didn’t say.
- Don’t show the product doing what it can’t. That’s a problem however the video was made.
- If you’re unsure, label it. It’s the cheaper mistake.
What we didn’t cover
We read New York because it’s the state rule we kept running into. Other states and other countries, including the EU, may have their own, and we haven’t checked them. Rules on likeness, meaning using a real person’s face or voice, are a separate matter, and a serious one. These pages change, so look again before a big campaign.
We wrote this because Lou by Luven AI, the video tool we’re building, has to get it right. Its free previews use no AI shots or AI voice, though AI writes their scripts. A paid video with AI shots or the Lou character in it is meant to carry an AI label wherever the platform offers one. Its rules bar AI people posing as customers. Lou isn’t open yet, none of this has run on real accounts, and no lawyer has checked this page yet.
Sources, read 3 and 4 Nov 2026
How we write these: with AI help, from the sources we link, with dates. Counts marked as ours were made by an AI reader with no second check. If something’s wrong, email everest@luven.ai and we’ll fix it and say so.
