The EU AI Act and AI-Generated Ads: What Brands Must Label

The EU AI Act advertising rules have applied since 2 August 2026. If an ad uses AI to show a realistic person who didn't actually do what the ad shows, such as an AI avatar "reviewing" your product, it must be clearly labelled as AI-generated from the first second. Videos of real creators filmed with their consent need no AI label, and neither do routine edits like colour correction or captions.

This article is general information for brands, not legal advice.

The rule in plain words

Article 50 of the AI Act says that whoever uses an AI tool to create a deepfake must disclose that the content is AI-generated or manipulated. If your team makes AI avatar ads in-house, that is you.

On 20 July 2026 the European Commission published guidelines on these transparency rules. The points that matter for advertisers:

• The rules apply from 2 August 2026. Deepfakes made before that date don't need labels added after the fact.
• The label must appear at the latest at first exposure, for example at the start of the video or when the ad appears in someone's feed.
• A hidden watermark is not enough. The label must be visible or audible.
• If you simply hire an agency and don't decide how AI is used, the agency is responsible for labelling, not you.
• Fines go up to €15 million or 3% of worldwide turnover, whichever is higher. For SMEs, whichever is lower.

The AI Omnibus amendment gives AI tool makers more time for their machine-readable marking duties (Paul, Weiss). That delay does not change the advertiser's duty to label a deepfake.

Do AI avatar ads count as deepfakes?

Usually, yes. A deepfake is AI content that looks like a real person, object, place or event and could pass as authentic. The guidelines say this includes realistic AI-generated human avatars, and the person only needs to look plausibly real. No intent to deceive is required.

The Commission's own examples read like a description of AI UGC. One is an "AI-manipulated video featuring a realistic synthetic influencer testing out a sponsored real product". The list also includes a teleshopping-style video of simulated consumers using a product to persuade viewers to buy. Only clearly artistic, satirical or fictional work gets a lighter regime, and it still needs a disclosure.

What doesn't need an AI label

Real creators filmed with consent don't need one, because there is no AI-generated person to disclose. Normal advertising rules still apply: ads must be recognisable as ads under EU consumer law, and a creator posting on their own channel must use the paid-partnership label their country requires.

Routine edits don't need one either. The guidelines treat colour correction, lighting, noise reduction and audio clean-up as minor changes. Auto-captions and translated subtitles of what the creator actually said are treated as standard editing.

A real product on an AI-generated background is fine as long as the ad doesn't mislead about the product. An AI image that makes the product look better than it really is counts as a deepfake.

AI dubbing is the grey area. The guidelines list voice cloning as a deepfake example and treat realistic speech in a specific person's voice as a substantial change. To be safe, label any AI dub of a real creator in their own cloned voice, and get their written consent first. Translated subtitles avoid the question entirely.

Do AI labels hurt ad performance?

The evidence says yes. In experiments by NIM (Nuremberg Institute for Market Decisions), labelling an ad as AI-generated made people less willing to research or buy the product. Consumers saw these ads as less natural and less useful.

Human creative wins even without labels. In May 2026, Ipsos and the S.I. Newhouse School tested 10 human-made ads against 10 AI recreations. Human ads were 14% stronger on short-term effectiveness and 17% stronger on long-term brand equity.

What could change next

The European Parliament lists the Digital Fairness Act as announced for Q4 2026, with no text yet. Influencer marketing is in scope, and the Commission's consultation asked whether brands and agencies should help ensure influencers comply (Osborne Clarke). If that becomes law, brands could share responsibility for creators' disclosures.

A short checklist for brands

1. List every ad that shows a person, voice or product made or altered by AI.
2. Label realistic AI people and cloned voices from the first second, not in the caption.
3. Keep AI edits minor: colour, sound, crops and captions are fine; face swaps and "improved" products are not.
4. Agree in writing with your agency who decides on AI use and who labels it.
5. Get a creator's written consent before any AI dub or edit of their face or voice.

For most small and mid-sized brands, real people filmed with your product is the simplest compliant option, and the research suggests it performs best too. Our comparison of UGC and AI video tools covers where AI still helps. On Noovid, vetted creators across the eurozone film your product from €75, with commercial usage rights included and no subscription.

FAQ

Do I need to label UGC videos made by real creators?

Not under the AI Act. A real person filmed with consent isn't a deepfake, though normal ad and paid-partnership disclosure rules still apply.

Are AI avatar ads illegal in the EU?

No. They are legal, but a realistic AI avatar must be clearly disclosed as AI-generated at first exposure.

Does AI colour grading or captioning need a label?

Generally not. Cloning a real person's voice is treated differently.

Sources

• Guidelines on transparency obligations for providers and deployers of AI systems, European Commission, July 2026: deepfake definition and examples, deployer role, minor-edit exceptions, first-exposure timing, penalties, 2 August 2026 application, no retroactive labelling.
• Regulation (EU) 2024/1689 (AI Act), EUR-Lex, 2024: Article 50(4) deployer disclosure duty and Article 3(60) deepfake definition.
• EU Finalises Transparency Rules for AI-Generated Content, Paul, Weiss, August 2026: deferrals for provider marking obligations.
• Transparency Without Trust, NIM Insights, 2025: effect of AI labels on ad evaluation and purchase willingness.
• AI Ads Are Good Enough, and That's the Problem, Ipsos, May 2026: human ads 14% and 17% stronger than AI recreations.
• Digital Fairness Act, European Parliament Legislative Train, September 2026: DFA status and Q4 2026 timing.
• Digital Fairness Act Unpacked: Social Media Influencers, Osborne Clarke, September 2025: consultation question on brand responsibility for influencer compliance.