AI UGC disclosure rules: what Meta, TikTok and YouTube actually require

What each platform actually requires

The confusion comes from treating this as one rule. It is three rules, written by three companies with different priorities, and they disagree about the basics.

TikTok has no commercial carve-out

TikTok asks creators to label all AI-generated content containing realistic images, audio or video. There is no exemption for ecommerce or product advertising. Its advertising policy says plainly that undisclosed AI-generated content gets rejected or restricted, and political AI content is banned from paid placement entirely, with no disclosure route around it.

Meta splits its rulebook in two

Meta reserves hard, written self-disclosure for ads about social issues, elections and politics. That policy does not extend to product advertising. A standard commercial AI UGC ad sits outside the mandatory-disclosure zone on Meta today.

That is not the same as invisible. Meta applies an AI Info label automatically where it detects generated content, using C2PA Content Credentials and its own classifiers, and gives you a manual control for cases it misses. You are not required to declare it. Meta may well label it anyway.

YouTube cares about realism, not category

YouTube asks a narrower question: could a viewer mistake this for something that actually happened? If yes, tick the altered or synthetic content box. If the content is obviously stylised or animated, no disclosure is needed.

What triggers disclosure, and what does not

The line that matters across all three platforms is whether AI generated or substantially altered a realistic person, voice or event. Production help does not count.

Disclose

• An AI presenter or avatar delivering the script
• A synthetic voice standing in for a real person
• A real person appearing to say or do something they did not
• Footage of a real place or event altered to show something that did not occur
• A face swap, or a digitally recreated likeness

Do not disclose

• AI used to write or refine the script
• AI captions, subtitles or translations
• AI-generated thumbnails and static graphics
• Colour grading, noise removal, background blur, upscaling
• Obviously unrealistic or cartoon-style AI visuals
• Voiceover laid over stock footage or illustration

How to label, platform by platform

TikTok

1. Use the AIGC toggle in the post or ad creation flow. It adds a visible "AI-generated" tag.
2. If you upload through Ads Manager, declare it in the creative settings rather than relying on the organic toggle.
3. Alternatively, burn your own clear caption, watermark or sticker into the creative. TikTok accepts self-labelling as long as it is legible.
4. If your export carries C2PA Content Credentials, TikTok will read them and label automatically.

Meta

1. For product ads, no action is required. Meta applies its AI Info tag where it detects generated content.
2. For political or social-issue ads, declare it in the disclosure field during ad creation. This is mandatory and enforced.
3. Use the manual AI disclosure control if you want the label applied to a product ad that Meta has not flagged.

YouTube

1. In YouTube Studio, open the video details page.
2. Under altered content, select yes if the video could be mistaken for real footage.
3. YouTube adds a "Modified or Synthetic" line to the description. For sensitive topics it appears on the video itself.

The whole process takes under a minute per asset. The cost of getting it wrong is considerably higher.

What happens if you skip it

Enforcement is escalating rather than binary. On TikTok, undisclosed AI creative gets rejected or restricted at review. On YouTube, creators who repeatedly fail to disclose face manually applied labels, content removal, or suspension from the Partner Programme.

The subtler penalty is worse. YouTube now proactively detects synthetic content and applies a disclosure label you cannot remove. That label signals to the platform that you tried to conceal it, and that signal follows the account rather than the individual video.

The EU AI Act and the FTC

Platform policy is one layer. Law is another, and it arrives shortly.

The EU AI Act requires disclosure of AI-generated content from August 2026. Unlike platform rules, this applies regardless of where the ad runs, and it carries statutory penalties rather than account restrictions. If you advertise to anyone in the EU, this reaches you.

In the United States, the FTC treats AI UGC the same way it treats human UGC on the question that matters most: you cannot fabricate a testimonial. An AI presenter saying "this changed my skin" is a fabricated first-person endorsement if no such person exists and no such result was measured. That is a deceptive-practices problem, not a labelling one, and no amount of AI disclosure fixes it.

The FTC also expects layered disclosure on paid content: one label identifying the advertisement, a separate one identifying the AI. They are not interchangeable.

Does labelling hurt performance?

This is the real question behind the compliance question, and the answer is reassuring. Early data on disclosed AI ads shows minimal effect on click-through rate across most product categories. Disclosure has not been shown to reduce performance in testing.

Which makes the calculation simple. You are weighing a negligible creative cost against ad rejection, account-level trust signals, and from August 2026, statutory liability. There is no version of that trade that favours hiding it.

Where AI creative genuinely underperforms, it is not because of the label. It is category fit. AI presenters hold up well in home goods, apps, apparel and food. They struggle in before-and-after transformations, where the viewer clocks the synthetic face and immediately assumes the result is synthetic too.