AI labels and platform rules
Lesson 10 of 16 in the course “Marketing 2027”. Free, with animations and a quiz.
AI labels are now law
Since August 2, 2026, Article 50 of the EU’s AI law, the EU AI Act, has been in force in the European Union. It requires ads where AI created a realistic person, voice or event to be clearly labeled. Importantly, the law isn’t just about European companies: if your AI ads reach people in the EU, the rules apply to you wherever you are. Responsibility for the label lies with whoever publishes the ad, meaning the brand or agency. Fines for transparency violations go up to €15 million or 3% of global turnover, and for small businesses the lower amount applies. Even if you’re outside the EU, platforms already apply similar rules to everyone.
What to label and what not to
The law targets content that looks real even though AI made it. You need a label if an AI avatar plays a customer, there’s a synthetic voice, the ad shows an event that never happened, or the product looks better than in real life. That last one matters a lot for clothing: fabric “improved” by AI is already deception. A real person filmed on camera doesn’t need a label. Nor does ordinary editing: color grading, noise removal, auto captions, background blur. Under the guidance to the law, the label must be noticeable right away, without a click. And ads almost never count as an “evidently artistic work”, which is where the exception applies.
How the platforms label AI
Platforms don’t wait for you to remember the label. Meta scans uploaded files for technical traces of AI and adds an “AI info” label, and for ads it shows a more detailed disclosure. Since July 2026 Google has been rolling out a “How this ad was made” panel that shows whether an ad was created or edited with AI, across Search, YouTube and Discover. Google adds the disclosure itself for its own generators, but advertisers must declare third-party AI tools themselves. TikTok automatically adds an “AI-generated” label when it finds technical markers and requires creators to declare AI themselves. If you don’t, the content can be removed.
C2PA: a passport inside the file
How do platforms know about AI? Many generators write metadata into the file using the C2PA standard. It’s like a passport: what made the file, when, and what edits were made. EU law separately requires AI tool makers to mark their outputs in a machine-readable way. Platforms read that passport and add the label themselves. So the takeaway is simple: trying to wipe metadata to hide AI is a bad idea. Platforms use other signals too, and if the deception comes out, trust in the brand takes the hit. It’s simpler and safer to declare AI honestly when uploading and keep the source files with their metadata.
Why real people are back in demand
All these rules lead to an unexpected conclusion: videos with real people gain more advantages. They don’t need an AI label. Viewers have fewer reasons to ask “is this even real?”. You can honestly show the test and the review in them, the strongest kinds of proof from the ad anatomy lesson. And emotions feel genuine; we saw with Coca-Cola how sharply people react when they sense a swap. That doesn’t mean AI is banned. It’s an argument for the hybrid: film the truth live, and let AI help with variations, formats and translation.
Decide in 3 questions
So you don’t have to keep the whole law in your head, ask three questions before every launch. First: is there a generated person, voice or event that never happened? If yes, label it. Second: does the product look better than in real life because AI “touched it up”? If yes, it needs a label, and better yet, redo the video, because that’s also deceiving the buyer. Third: not sure? Then label it too. An honest label costs almost nothing, while a dispute with a platform or regulator is expensive. Only if all three answers are “no” is no label needed. Remember: this is a learning framework, not legal advice, and rules differ by country.
Do it yourself: a 20-minute label audit
Open your ad account and list all active creatives. Run each through the three decision questions. Wherever even one answer is “yes”, add a label that’s visible without a click: on-screen text or the platform’s built-in disclosure. When uploading, use the toggle or field where the platform asks you to declare AI, instead of hoping it won’t notice. Save the source files with metadata in a separate folder: if a dispute comes up, you’ll have proof of how and with what the video was made. That’s twenty minutes after which neither reviews nor customer questions scare you. Next up: tests without fooling yourself.
Check yourself
- Does it need an AI label?
Correct answer: Yes: an AI avatar is playing a customer. A realistic generated person who looks like a real buyer is exactly the case the label exists for. - Does this one need a label?
Correct answer: No: it’s ordinary editing, the person and hoodie are real. Auto captions and color grading are assistive editing; they don’t create anything new. A real person on camera needs no label. - What should she do with the metadata?
Correct answer: Keep them and honestly declare AI when uploading. An honest declaration removes the risk of disputes and removal. Source files with metadata are proof if questions come up. - Does the EU law apply to Nika?
Correct answer: Yes, if her AI ads reach people in the EU. The law looks at where people see the content, not where the business is registered. Show AI ads in the EU and you follow the rules.
← Making creatives with AI Testing without fooling yourself →