AI UGC disclosure by platform
UGC-style ads with a realistic generated person need platform AI labels and, for EU audiences, an Article 50 first-exposure mark.
A UGC-style ad that uses a realistic generated person is still an ad, and it is still synthetic media. TikTok, YouTube, and Meta each want their own AI label on that file. Article 50 of the EU AI Act, in force since 2 August 2026, wants a visible or audible mark at first exposure when an EU audience would take a realistic person, place, or event as authentic. The paste-ready sentences and the three EU icons live in the Article 50 disclosure template. This page is which switch you flip for UGC-shaped content, and why "it looks like a customer" is not an exemption.
AI UGC is brand-owned, feed-native video generated to look like a person talking to camera about a product. It is not an unprompted customer post. Treating it as organic UGC is how teams skip the label and then argue with a takedown. 48% of UGC platforms launched 2023–25 are AI-native (archive.com; creator.co). The format is common. The disclosure is still required.
The label is not a ranking penalty on the major platforms. Unoriginal, templated, mass-produced uploads are. That split is documented in the AI label is not a reach penalty. This page does not re-litigate it. It maps the UGC ad case.
What counts as UGC-shaped synthetic media
If a reasonable scroller would think a real person used the product on camera, and that person was generated, cloned, or substantially altered, you are in the disclosure bucket. Bathroom lighting does not move you out. A "not a paid actor" caption does not move you out. A real product in the overlay does not move you out.
In:
- Generated or avatar presenter, realistic, talking to camera.
- Digital twin of a real creator, even if the creator licensed the twin.
- Face or voice swap onto a real body.
- Fully generated scene that still looks like a customer kitchen.
Out, or at least not this page:
- Real creator, real face, real voice, paid to deliver UGC. That is a paid-partnership / endorsement disclosure (FTC in the US, local advertising rules elsewhere), not an AI label. You still need the paid label. You do not need the AI label because of AI.
- Assistive edits: captions, colour, crop, denoise, translation burned as text. Standard assistive editing is the Article 50 exemption the Commission describes. Captions are not a deepfake.
- Obvious fiction and satire, labelled in a way that does not spoil the work, under the artistic carve-out. A UGC ad that is trying to look like a real review is not satire.
When in doubt on a realistic person, label. The cost of the badge is lower than the cost of a Creativity Program suspension or an Article 50 problem. Human UGC still rates 81% on authenticity versus 63% for AI UGC (Kapwing; Billo; creator.co). Hiding the generate will not close that gap. It adds a deception problem on top of it.
Paid-partnership language and AI language stack. "Paid partnership with Brand" does not tell a viewer the face is synthetic. "AI-generated" does not tell a viewer the brand paid. Put both on when both are true. UGC style is not a third category that skips either.
TikTok, YouTube, Meta: the UGC-ad version
TikTok. Use the built-in AI label. TikTok has said AI content remains eligible for the Creativity Program when that label is used, at the same rate; undisclosed AI can mean removal or program suspension (socialscalehub, 2026 disclosure roundup; vexub; fliki). For a UGC-style ad, that means: label on, even if you are running Spark Ads or a code. Spark does not launder provenance. If you are running ads through Ads Manager rather than boosting a post, use the advertiser disclosure path Meta-and-TikTok ads already require, and the AI label if the creative is synthetic. Do not rely on a caption that says "POV" as a label. POV is a format.
YouTube. Disclose realistic synthetic media of real people through the altered-content flow. YouTube's inauthentic-content policy is the one that hits mass-produced, repetitive, templated uploads — not AI as such. AI content stays monetizable when it carries the creator's own vision and the required disclosure (same 2026 policy roundups). A farm of near-identical UGC avatars reading the same script is an inauthentic-content problem even if every file is labelled. A labelled, specific UGC ad that is actually yours is not that farm. Shorts and long-form both get the disclosure; Shorts UI is tighter, so do not hide it only in a description a Shorts viewer never opens.
Meta (Feed, Reels, Stories, ads). Labels are required. Meta relies on C2PA/IPTC signals and self-disclosure; Reels bonuses depend on disclosure (same roundups). For UGC-style ads, self-disclose in the ad setup and keep Content Credentials on the file if your pipeline writes them. Stripping C2PA to "look more native" is the wrong optimisation. The badge is not the documented reach penalty. A realistic generated endorser without a label is.
Ads versus organic is not an exemption on any of the three. A dark post is still a publication. A Spark ad is still a publication. An Advantage+ catalogue video that uses a generated talking head is still a publication. If the creative would need a label as a Reels post, it needs a label as an ad.
Symphony Agent and other platform-native generators may apply invisible watermarks and labels for you when you make the file inside that tool. Files you bring from elsewhere, including a UGC generator, do not inherit that. Assume you owe the label on ingest.
Article 50 on a UGC-style ad
Article 50 is a law, not a platform toggle. Commission guidelines were adopted 20 July 2026; the transparency duties apply from 2 August 2026 (EC FAQ; Bird & Bird; Morgan Lewis).
For a UGC-style ad the relevant fork is usually deployer publishing a deepfake-shaped asset: content resembling a real person that would appear authentic. The duty is a visible or audible label at first exposure. A description field, a landing page, or a "see more" tap is not first exposure. First frame, or a spoken line at the start, or an on-screen line that is on screen when the file starts.
If you are the company that built the generator, you also have provider duties: machine-readable marking of synthetic audio, image, video, and text, with a grace period to 2 December 2026 for systems already on the market before 2 August 2026, and watermark-detection interoperability running to 2 February 2027. Most brands running UGC ads are deployers, not providers. Do not skip the visible label because the model vendor watermarked the file. Viewers cannot see a watermark.
Fictional-looking ads may use the artistic carve-out and label in a way that does not spoil the work. A UGC ad whose entire value is "this could be a customer" is not that carve-out. Label it as generated.
EU icons exist (fully AI-generated, AI-modified, basic icon with a second layer). Fines can reach €15 million or 3% of worldwide turnover (Paul Weiss; CSA). This is not legal advice. It is the reason a bathroom-style ad still gets a first-frame mark if EU viewers will see a realistic person.
Content generated and published before 2 August 2026 is not retroactive. Generated before and published after is covered. Recutting an old unlabelled generate for a new EU campaign is a new publication.
Non-EU brands still hit this when the audience is in the EU. Geo-targeting "US only" is an operational control, not a legal memo. If the file can be viewed in the EU, plan for the label.
Put the label in the edit, then tick the platform box
Platform toggles die on download, on a recut, on a Spark that points at a different asset, on an Amazon recut, on a cutdown. Burn a short line into the file: "This video is AI-generated" or "This person is AI-generated," small, first seconds, in a safe area. Then also tick TikTok, YouTube, and Meta. Two layers. One file that travels.
Do not hide the line in a joke. Witty labels fail first-exposure tests. Do not put it only on the last frame. Do not put it only in a comment. Do not skip the words because the format is casual.
If a real creator licensed a digital twin, say both: paid partnership, and synthetic performance. Viewers who know the person are the ones most likely to feel tricked by a twin.
Production-wise, treat disclosure as a caption-style overlay in the same pass as burned-in captions. The UGC video generator can assemble the talking overlay; the label is a text layer you ask for on every synthetic presenter, not a once-per-campaign design. For the exact paste lines and icon placement, use the template. For the UGC decision, use this page: if it could pass as a customer and it was generated, label it on every surface it will run.
Sign up if you want that overlay in the same edit as the captions, not as a Friday panic.
FAQ
Does a UGC-style ad need Article 50 if I only run it on Meta in the US?
Article 50 follows the audience and the content, not the ads manager dropdown. A US-only campaign that cannot be viewed in the EU is a narrower fact pattern than a public Reel. If EU viewers will see a realistic generated person, plan for first-exposure labelling. This is not legal advice; it is the conservative production default.
If TikTok labelled it for me, am I done?
You are done with TikTok's toggle for that upload. You are not done with YouTube, Meta, ads recuts, or Article 50. Burn a line into the file so the next surface inherits it.
Can I skip the AI label if I already say "paid partnership"?
No. Paid partnership discloses the commercial relationship. The AI label discloses that the person or performance is synthetic. A paid generated presenter needs both. A paid real creator needs the partnership disclosure and does not need an AI label because of AI.