Labeling AI Advertising: What EU AI Act Article 50 Changes for Brands

The EU AI Act's transparency rules (Article 50) have been in force since August 2, 2026, and the Commission published its final implementation guidance on July 20, 2026. But 'all AI ads must be labeled' is an oversimplification. This article covers who, when, and for which content the labeling duty actually applies — and why the guidance isn't legally binding.

• The EU AI Act's transparency obligations (Article 50) have been in force since August 2, 2026.
• 'All AI ads must be labeled' is an oversimplification — the obligation depends on whether you're a provider or deployer, and on the content type (Article 50, paragraphs 1-5).
• The clearest scope: visual or audio content that resembles or impersonates a real person (deepfake-type content) — labeling is required regardless of intent, with narrow exceptions.
• The European Commission published its final (not draft) implementation guidelines on July 20, 2026 — but that guidance is still non-binding; only the Court of Justice of the EU (CJEU) can give an authoritative interpretation of the law itself.

The EU AI Act's (Regulation (EU) 2024/1689) transparency rules — Article 50 — have been in force since August 2, 2026, and the European Commission published its final implementation guidance on July 20, 2026. Despite that, the version circulating in most conversations is 'all AI advertising now needs a label.' The real scope is narrower and more technical. For any brand advertising into the EU market or exporting there, understanding that gap matters — both to avoid unnecessary panic and to not miss a real obligation.

What Article 50 actually says

Article 50 defines five separate obligations, split between the provider (the company building the AI system) and the deployer (the company using it — a brand or agency, for instance). On the provider side: AI systems that interact directly with people must make that clear, and systems generating synthetic audio, image, video or text must mark their output in a machine-readable format. On the deployer side: people exposed to emotion-recognition or biometric categorization systems must be informed, deepfake content and AI-generated public-interest text produced without editorial oversight must be labeled, and all of this information must be given clearly at the point of first interaction. For a brand or agency, the obligation that matters most in practice is the deepfake-labeling duty.

Who this applies to

The obligation covers providers and deployers operating in the EU, or publishing content/advertising aimed at the EU market. For a Turkey-based brand or agency, that means it's directly relevant whenever you're exporting to the EU or targeting EU audiences — not only for companies legally established there.

The deepfake rule — the clearest part of the scope

Article 50(4) applies to AI-generated or manipulated content that resembles a real person, object, place, entity or event in appearance or sound — deepfake-type content. The deployer is obligated to clearly disclose that such content was artificially generated or altered; the text of the law doesn't require any intent to deceive for this duty to apply. For content that's obviously artistic, creative, satirical or fictional, the obligation is lighter: it's enough to disclose the content's existence in a way that doesn't interfere with the work's display or enjoyment. Uses authorized by law for criminal investigation or prosecution are exempt.

Why 'every AI ad needs a label' is wrong

Because the law itself carves out exemptions. The marking obligation under Article 50(2) doesn't apply when a system only performs an assistive function for standard editing and doesn't substantially alter the input data. The public-interest-text labeling duty under Article 50(4) doesn't apply when the content has gone through human review and a natural or legal person holds editorial responsibility for publishing it. In practice, the scope centers on two things: deepfake-type content impersonating a real person or event, and AI-generated public-interest text published without editorial oversight. Everyday AI-generated ad copy, product descriptions or social posts fall largely outside this obligation.

Timeline and the status of the guidance

One important nuance: the Commission's July 20, 2026 guidance is now final rather than a draft, but it still isn't legally binding — it's soft law. Only the Court of Justice of the European Union (CJEU) can give a final, authoritative interpretation of how the law itself applies. So the practical interpretations you're reading today, including this one, reflect the best current understanding based on the Commission's official guidance and existing legal commentary — not a fixed, unchangeable reading of the statute.

What brands should do

If a campaign aimed at the EU uses AI content that resembles or impersonates a real person, build disclosure into your standard production workflow rather than treating it as an afterthought. Keep in mind that saying 'a human reviewed it' isn't by itself a sufficient exemption — it requires genuine editorial oversight and accountability. If you're exporting into the EU, talk to legal or compliance counsel to pin down your exact obligation; this article isn't a substitute for that. And keep an eye on both the Commission's July 20, 2026 guidance and any future CJEU rulings, since the latter is what will eventually settle the open questions.

Note: this article is for general information only, not legal advice. We recommend consulting a legal or compliance advisor to determine your specific obligations for your own campaigns.

Want to talk through the production process for your AI-assisted campaign aimed at the EU market?

Blog · Get a quote