A finished AI video ad now has one more job before it goes live: someone has to decide whether it needs a label, and who adds it. In most projects nobody owns that decision until the ad is already scheduled.
This is a plain-language guide for brands and agencies. It is not legal advice, and the rules differ by country, platform and ad type. What it gives you is the list of questions to settle with your studio before delivery, so a label never becomes a last-minute scramble.
What does the EU AI Act actually require?
The transparency rules sit in Article 50 of the EU AI Act, and they apply from 2 August 2026. Two parts matter for video.
The first is a duty for the people who use an AI system to make a deepfake. The text says deployers "shall disclose that the content has been artificially generated or manipulated" (Article 50, AI Act text). If the content is part of an evidently artistic, creative, satirical or fictional work, the duty is lighter: the disclosure must be made "in an appropriate manner that does not hamper the display or enjoyment of the work." In ad terms, that means a label can be quiet, but it cannot be missing.
The second is a duty for the providers of the generator: outputs must be marked in a machine-readable way so tools can detect them as AI-made. That part is the tool vendor's job, not yours. Law-firm summaries note that the recent Digital Omnibus gives systems already on the market a grace period for this marking until 2 December 2026, while the other Article 50 duties proceed as scheduled (Gibson Dunn).
Two honest limits. We read the Article 50 text, but the Omnibus details come from law-firm commentary, not the Official Journal. And whether a particular ad counts as a "deepfake" depends on how realistic it is and what it shows. Ask your lawyer about the edge cases.
Do platforms add their own rules?
Yes, and they change often. Based on trade coverage and ad-tech blogs, not the platforms' own pages, which we could not open reliably:
- TikTok is described as the strictest: AI-generated ad content is allowed with its AI-generated content label or a clear disclaimer, caption or sticker from the advertiser.
- Meta is described as requiring disclosure for ads about elections, politics and social issues, with a "generative AI used" label appearing on some ads.
- YouTube is described as asking creators to label realistic synthetic content, with firmer rules for political ads and no blanket mandate for ordinary commercial ads.
Treat that list as a prompt to check, not as a rule book. Open each platform's current ad policy for the placements you are buying, and note the date you checked.
Who is responsible: the brand, the agency or the studio?
In practice the advertiser is the one whose name is on the ad, so the brand carries the risk. That does not mean the studio can ignore it. A studio knows exactly what was generated, what was filmed and what was edited, and that is the information a disclosure decision needs.
We suggest putting it in writing at the start of a project:
- The brand decides the disclosure policy for its channels and markets.
- The studio documents which parts of each deliverable are AI-generated or AI-altered.
- The studio delivers the version with the label burned in, and a clean version without it, clearly named.
- The brand, or its agency, confirms placement rules and signs off the labeled version.
What should the studio deliver?
When we hand over an AI video, these items make the labeling step easy:
- A labeled master and a clean master, in every aspect ratio you ordered.
- A short production note: which tool made which shots, and what was real footage.
- A caption line you can reuse, such as "Created with AI assistance," in each language you run.
- The generation records for each shot, kept in case a platform or a client asks. Keep them for as long as the ad runs, because questions tend to arrive months after delivery.
If you commission batches of short ads, the same structure applies. Our guide to briefing AI UGC-style video ads shows what to put in the brief up front, and disclosure belongs on that list.
Does a label hurt performance?
We have no data to claim either way, and we would be wary of anyone who does. What we can say is that a visible label is a smaller risk than a rejected ad, a removed post or a customer who feels tricked. Where the work is clearly stylized, a short, calm label usually sits fine with the creative.
The practical takeaway
Before an AI video ad is approved, ask five questions:
- Does it show a realistic person, place or event that did not happen?
- Which markets and platforms will it run in, and what are their current rules?
- Who decides the label wording, and who adds it?
- Do we have a labeled version and a clean version of every format?
- Where are the generation records kept?
If the answer to any of these is "we will sort it out later," that is the one to fix first. Licensing is a separate question, and we cover it in Commercial Use of AI-Generated Images. To see how we approach finished video work, visit 35milimetre.com.