AI labelling in advertising
By Syntax Motion, Vienna. Updated: 16 October 2026
The transparency duties in Article 50 of the EU AI Act have applied since 2 August 2026. For advertising that means: if an AI spot looks convincingly real and shows people, places or events that never existed in that form, it has to be disclosed as artificially generated. For obviously creative or fictional work a restrained note is enough.
General information, not legal advice. Every statement here is checked against the primary sources linked below, as of 9 October 2026. For your project, the contract we sign with you is what counts.
What the AI Act requires
Article 50 of Regulation (EU) 2024/1689 distinguishes two roles, and with advertising films both can be involved.
The duties have applied since 2 August 2026. A transition period exists only for providers of AI tools that were already on the market before that date: they have to implement the machine readable marking (paragraph 2) from 2 December 2026. For you as the advertiser that changes nothing about disclosure (paragraph 4). Source: European Commission, FAQ on Article 50.
| Role | Who that is | Duty |
|---|---|---|
| Provider (Art. 50(2)) | makers of the AI tools, for video or image generation for instance | mark outputs in machine readable form so they can be recognised as artificially generated |
| Deployer (Art. 50(4)) | whoever uses an AI system under their own responsibility, so the agency and, depending on the setup, the advertiser | disclose that content was artificially generated or manipulated, where it is a deepfake |
When is an advertising spot a deepfake?
The regulation defines a deepfake as AI-generated or manipulated image, audio or video content resembling real persons, objects, places, entities or events, and which would falsely appear authentic to a person (Art. 3(60)).
- Probably yes: a photorealistic spot with a person who reads as a real customer, or a product shown in a convincing everyday scene that never took place.
- Probably not, or only in part: obviously fantastical worlds, animated characters, scenes that clearly could not be real.
The European Commission names three features that all have to apply: the content strongly resembles a person, object, place or event; that thing exists or could plausibly exist; and the content could falsely appear authentic to the audience. AI-generated backgrounds and visual effects in ordinary film production do not, as a rule, appear convincingly authentic (European Commission, FAQ on Article 50). The Austrian Economic Chamber adds that recognisably unrealistic depictions, ones that contradict the laws of physics for example, are not deepfakes and need no label (WKO).
As a rule of thumb: the more easily a scene could be taken for a recording of reality, the more likely it has to be labelled. Where that is unclear in a specific case we label anyway, because a line in the description costs nothing and ends the risk.
The exception for creative works
Where content forms part of an evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing the existence of generated content in an appropriate manner that does not hamper the display or enjoyment of the work (Art. 50(4)). Many commercials with obviously staged, heightened imagery are likely to fall under this, although whether advertising counts as an evidently creative work is not finally settled. A short, discreet note is then enough.
How to label without the spot suffering
- An on-screen line at the end or in the credits, "Made with AI" for instance, clearly legible but not over the main scene.
- A note in the accompanying text of the ad or post, in addition to the on-screen line.
- Keep the metadata: many tools write provenance information into the file. Do not strip it on export. On its own it is not enough: according to the European Commission, deployers have to disclose visibly or audibly and cannot rely on the tool's machine readable marking.
- Use the platform declaration: YouTube, Meta and TikTok require their own declarations for realistic AI content. The rules are at YouTube (disclosing altered or synthetic content), TikTok (AI-generated content) and Meta (AI info), retrieved 9 October 2026. These rules apply independently of the statute and are enforced by the platforms themselves, usually through reduced reach rather than a fine.
- Document it: which tools were used and which scenes are generated.
Who is liable, the agency or the advertiser?
That depends on who deploys the AI system under their own responsibility and who decides on publication. In practice agency and advertiser should put in the contract who labels and in what form. We set it out in clause 6.3 of our terms: we supply the notice, and responsibility for publication and the labelling required there stays with the client.
What penalties apply
The AI Act provides for fines of up to 15 million euros or 3 % of worldwide annual turnover for breaches of the transparency duties, whichever is higher; for small and medium sized companies the lower amount applies (Art. 99). Independently of that, unlabelled convincing advertising can also be attacked as misleading under unfair competition law.
How we handle it at Syntax Motion
Every film leaves here with three things: the provenance note in the file metadata, a ready written line for the video description, and a note on which scenes are generated. On request we put the notice on screen, discreetly in the credits or in the first second. In the first call we say whether your project falls into the scope of the disclosure duty at all. For spots deliberately designed to be unrealistic, it usually does not. For client stories and customer voices the stricter rules of testimonial video and AI apply.
Frequently asked questions
Does every AI advertisement have to be labelled?
Not every AI video has to be labelled. Since 2 August 2026, Article 50 of the AI Act requires disclosure for deepfakes, meaning convincing content showing real or plausibly real people, places or events. YouTube, Meta and TikTok have their own rules on top. For obviously creative or fictional work a restrained note is enough.
Since when has the labelling duty applied?
The transparency duties in Article 50 of the AI Act have applied since 2 August 2026.
Does the EU AI Act apply to us if we are not in the EU?
It applies to content placed on the EU market. Advertising aimed at audiences in the EU is covered regardless of where the advertiser is based.
Is a note in the small print enough?
The information has to be clear and distinguishable and given at the latest at the first interaction with the content (Art. 50(5)). A hidden note is unlikely to meet that.
Read on: All guides • Copyright in AI video • Synthetic media
An AI spot, cleanly labelled
We plan the labelling in from the start, so that it fits the spot.
Book a first call