AI-Generated UGC Ads: 6 Checks Before You Deliver a Creative to Your Client (EU AI Act, TikTok, Meta)

November 6, 2026

· By Tarek Smiri

AI-Generated UGC Ads: 6 Checks Before You Deliver a Creative to Your Client (EU AI Act, TikTok, Meta)

Short answer

Since 2 August 2026, Article 50 of the EU AI Act requires whoever deploys a realistic AI-generated person, place or event in an ad to disclose it clearly. TikTok and Meta label AI content on their own terms, and those terms are broader than the law in some places and narrower in others. In a white-label chain, the harder question is often not whether to disclose but who does it. Run six checks before you hand a creative to your client: AI level, realism, script, product fidelity, labels, and who owns the disclosure. This is an operational guide, not legal advice.

Three rulebooks that don’t line up

An AI-generated UGC ad running in Europe answers to three sets of rules at once. They don’t use the same trigger.

Rulebook	What triggers it	What it asks for
EU AI Act, Article 50	A “deepfake”: AI content that resembles real people, objects, places or events and would falsely appear authentic	A clear, visible or audible disclosure by the deployer, at first exposure at the latest
TikTok ad policy	Content that is completely AI-generated or significantly modified by AI	TikTok’s AIGC label, or your own clear disclaimer, caption, watermark or sticker
Meta	Ad images created or significantly edited with Meta’s AI features, or created or edited with third-party AI tools	An “AI info” label that Meta applies, mostly in the “About this ad” menu

The gap matters in both directions. A stylised, obviously fictional AI ad may not be a deepfake under the AI Act, yet TikTok still expects a label because it is fully AI-generated, and an undisclosed AI ad there will be “rejected or restricted” (TikTok). On the other hand, Meta’s “AI info” label usually sits behind a menu. The European Commission says a deepfake disclosure must be perceivable without any dedicated action, and that deployers cannot rely on the machine-readable marking embedded by the tool (European Commission). A platform label is therefore a floor, not proof of compliance.

Who is the deployer in a white-label chain?

A typical chain has a production studio, an agency that delivers under its own brand, and a brand or media agency that runs the ad. Each one tends to assume the disclosure is someone else’s job.

The AI Act defines a deployer as anyone using an AI system under their authority in a professional capacity. The Commission adds that a legal person remains the deployer even when contractors or freelancers operate the system on its behalf and under its control. The French advertising regulator ARPP goes further for advertising. In its 3 August 2026 guidance, it says deployer status depends on who decides to use AI and how the output will be used or distributed. Depending on the situation, that can be the advertiser, the agency or another operator, inside or outside the EU, as long as the content runs in the Union (ARPP).

In practice: outsourcing the production does not outsource the decision. Write it down before delivery. That is check 6 below.

The six checks

1. Measure the level of AI in the final cut

Only what runs counts. Using AI for ideation, scripting, storyboards, translation or technical retouching does not, in principle, call for a disclosure. A UGC ad with a generated presenter, voice and set is at the high end: full or near-full generation. Record which elements are generated: face, voice, hands, background, product.

2. Assess realism and resemblance

The Commission’s three cumulative criteria for a deepfake are resemblance, a subject that exists or could plausibly exist, and a false appearance of authenticity. The last one depends on context and audience expectations. A photorealistic presenter in a believable kitchen meets the first two easily. A mascot, a stylised world or a fantastical character usually does not. If the presenter resembles a real person, a label is not enough: you also need that person’s prior consent and must respect data-protection rules.

3. Keep lived experience out of the script

This is the check that protects your client the most. A synthetic presenter saying “I’ve used this every day for three months” claims an experience that never happened. Since Directive (EU) 2019/2161, EU consumer law blacklists submitting or commissioning false consumer reviews or endorsements (EUR-Lex). The ARPP puts it plainly: misleading content stays misleading even with an AI label. Write demonstrations, explainers and problem-solution scripts in which the presenter shows and explains rather than testifies.

4. Check product fidelity, shot by shot

Generation drifts. Colours shift, textures improve, a bottle grows between two shots. The ARPP lists a product shown as more attractive or of better quality than it really is among the cases that require a label, and warns that exaggerated performance or features can amount to a misleading practice. Compare every shot with the real product or its official photos before you think about labels at all. TikTok also bans before-and-after comparisons that give a distorted impression of results.

5. Prepare the platform label and the on-screen disclosure

Switch on TikTok’s AIGC label, or add your own disclaimer, for fully generated or significantly modified content. TikTok advises labelling when in doubt. On Meta, expect the “AI info” label to appear on ads made with third-party AI tools (Meta). Meta’s Transparency Center also describes automated detection of third-party generative AI in ad media from 1 June 2026, in its policy for social-issue ads (Meta Transparency Center). Assume an undisclosed synthetic presenter can be spotted. When your analysis says a disclosure is required, put it in the file or the caption, visible from the start of the video, in the campaign’s language.

6. Decide who applies the disclosure and who keeps the record

Put three lines in the purchase order or the approval note:

Who decides whether a disclosure is required, and on what analysis.
Who applies it: burned into the file, added as a caption, or switched on in the ads manager.
Who archives the evidence: the version that ran, the analysis, the date, the platform, image and voice consents.

At Genesis Digital Factory, every ad ships with a one-page compliance report. It states whether an AI disclosure is required under Article 50 and why. It also records the models used, machine-readable marking, image and voice authorisations, third-party brands on screen and platform specs. That gives the agency a written record to pass on. The decision to run the ad and the campaign settings stay with whoever runs it, which is exactly what the approval note should say.

The evidence pack, at a glance

Check	The question	What to keep
1. AI level	Low, intermediate or high in the final cut?	List of generated elements
2. Realism	Could a person, place or product pass as real?	Deepfake analysis, consents
3. Script	Does a synthetic presenter claim a lived experience?	Approved script, final cut
4. Product	Is the product identical to the real one?	Shot-by-shot comparison
5. Labels	Platform label on? Disclosure visible from the start?	Screenshots of settings and file
6. Ownership	Who decides, applies and archives?	Signed approval note

If the campaign runs in France

Three French layers sit on top of the EU rules:

ARPP guidance (3 August 2026). A three-part grid covering the level of AI intervention, realism and impact on how consumers perceive the ad. No single criterion is enough on its own. The ARPP notes that the lighter regime for artistic or fictional works comes with “no exception for advertising”. It suggests indicative wording such as “Vidéo générée par IA” or “Personnage généré par IA”, shown from the start of a video.
The Conseil de l’Éthique Publicitaire (CEP). According to CB News, the CEP, a body associated with the ARPP, argued in July 2026 that a blanket “made with AI” label is unjustified. It locates the real risk where the person on screen is itself the brand’s promise, or where the image itself is a commitment, as in travel (CB News). UGC presenters often fall into the first case.
“Images virtuelles”. Under Article 5 of French law 2023-451, in force in its current form since 8 November 2024, content from paid commercial influencers that includes AI-produced faces or bodies must carry the words “Images virtuelles”. Breaches are punishable by one year’s imprisonment and a €4,500 fine (Légifrance). This matters when an AI ad is posted through a creator’s account.

Dates and stakes

2 August 2026: Article 50 applies. Content generated before that date does not need to be labelled retroactively, though the Commission encourages it.
2 December 2026: the transition ends for certain providers of generative tools already on the market before 2 August, for the machine-readable marking in Article 50(2) (AI Act Service Desk). That obligation sits with tool providers. It doesn’t replace the visible disclosure.
Fines: up to €15 million or 3% of worldwide annual turnover, according to the Commission, with proportionality for SMEs. Enforcement is mainly national.
Switzerland: no general AI law yet. A consultation on a federal bill is planned for spring 2027 (OFCOM). A Swiss agency whose ad runs in the EU should still apply these checks.

How Genesis Digital Factory fits in

We produce finished AI ads for European agencies, in white label and under your brand: UGC, product videos, image ads and, on request, cinematic spots. Human creative direction signs off the script before anything is generated. Editors finish every ad, and quality control checks product fidelity and AI compliance before delivery. A synthetic presenter never claims a personal experience as lived. Every project is quoted at a fixed price within 48 hours, with kick-off within 72 hours. The same quality control has stood behind more than 85,000 projects since 2005, from our production studios in Tunisia.

If the ad points to a landing page, the same contract can cover the page through our website production, so the page keeps the ad’s promise. For campaign management, see online advertising. For the rest of our white-label services, see white-label production.

Get a quote for AI ad production

Prefer to test first? Start with a trial project on a single creative.

FAQ

Does every AI-generated UGC ad need an AI label?

Not under the AI Act alone. The trigger is content that could falsely appear authentic, and a realistic synthetic presenter often meets it. TikTok, however, expects a label for any fully AI-generated ad content, so on that platform the practical answer is usually yes.

Is Meta’s “AI info” label enough to comply with Article 50?

Don’t count on it. The label often sits in the “About this ad” menu, while the Commission expects a deepfake disclosure people can perceive without taking any action. Add a visible disclosure when your analysis says one is required.

Can the agency push the responsibility onto the production studio?

Not by default. The Commission says a legal person stays the deployer when contractors work on its behalf and under its control. The ARPP says the role follows decision-making power over the use of AI and distribution. Agree in writing who decides, applies and archives.

Can a synthetic presenter say they love the product?

Avoid it. A first-person endorsement from someone who does not exist is a false consumer endorsement, which EU consumer law prohibits in all circumstances. An AI label doesn’t fix that.