Your teams use AI to produce content — articles, ingredient pages, posts, newsletters. Everyone does it; nobody writes it down. Since 2 August 2026, the European AI regulation (Regulation (EU) 2024/1689, the “AI Act”) has made the question official: certain AI-generated texts must now be labelled as such. The good news fits in one sentence — an exception tailored for brands that take their editorial work seriously. Provided you meet its conditions, and can prove it.
What Article 50 actually says (and who it targets)
The AI Act, adopted on 13 June 2024 and published in the EU Official Journal on 12 July 2024, is mostly known for its prohibitions and “high-risk systems”. But the chapter that directly concerns marketing and communication teams sits elsewhere: Article 50, “Transparency obligations”, applicable since 2 August 2026.
Its paragraph 4 addresses content producers in black and white: deployers of an AI system “that generates or manipulates text which is published with the purpose of informing the public on matters of public interest” must disclose that the text has been artificially generated or manipulated.
Health and nutrition tick the “public interest” box beyond debate: this is precisely the kind of subject for which the legislator wanted transparency. A brand’s health blog, an advice page, a dossier on an ingredient’s benefits — all of it informs the public on matters of public interest.
The exception that changes everything: human review and editorial responsibility
The same paragraph provides the way out — quoting the text:
“This obligation shall not apply […] where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content.”
In other words: the AI Act does not penalise using AI — it penalises publishing without control. The text borrows the logic of press law — a publication director, a review process, assumed responsibility — and extends it to brand content.
Two cumulative conditions:
| Condition | What it means in practice |
|---|---|
| Human review or editorial control | A real review process: source verification, scientific accuracy, claim compliance — not a copy-paste of AI output |
| Editorial responsibility assumed | An identified natural or legal person who signs: named author, publication director, statement in the legal pages |
If either is missing, the labelling obligation applies. And the bill can be steep: Article 99(4) provides, for breaches of Article 50, administrative fines of up to EUR 15,000,000 or 3% of total worldwide annual turnover, whichever is higher — capped at the lower amount for SMEs.
What this changes for a health-nutrition brand
Three profiles emerge across the sector.
Industrialised content without review. Product sheets, SEO articles and posts generated at volume, published as-is, no identified author. This is the use case Article 50 targets head-on: mandatory labelling — with the commercial effect you can imagine for health content stamped “AI-generated”.
AI-assisted content, controlled and signed. AI accelerates research and first drafts; a competent human verifies the cited studies, adjusts claim-safe wording, signs. The exception applies. It is the model of any serious newsroom — and the one we practise.
The in-between — majority and fragile. AI a bit everywhere, review sometimes, no trace of the process, no designated editorial owner. If questioned, nothing documents the right to the exception. Less a legal problem than an organisational one — and quick to fix.
Three reflexes to put in place now
- Designate and display editorial responsibility. A publication director in the legal notice, identified authors on content, a transparency note about the process. Cost: one hour. That is half the exception.
- Formalise editorial control. Who reviews what, against which criteria (sources, accuracy, claims), with what record? A written process, however short, documents the “human review” the text requires.
- Deal with the rest of Article 50. A chatbot on your site must disclose that visitors are talking to an AI (paragraph 1). Generated images and videos constituting deepfakes must be flagged (paragraph 4, first subparagraph). The audit takes a morning.
One detail that matters: these transparency obligations have applied since 2 August 2026. This is no longer a deadline to anticipate — it is the law in force.
Why this is an opportunity (not just a constraint)
Article 50’s paradox: it makes legally valuable what was already commercially valuable. Verified, sourced content signed by an identifiable expert was already what search engines reward, what AI answer engines cite, and what your B2B customers expect from a health brand. The regulation now makes it the boundary between content that must carry an “AI” label and content that needs none.
Brands that invested in a real editorial chain have nothing to change — only to document. The others now have a regulatory reason to do what their visibility already demanded.
FAQ
Is using AI for writing now forbidden or risky? No. The regulation does not prohibit AI-assisted writing; it requires transparency when content goes out to inform the public without human control. With editorial review and assumed responsibility, no label is required.
Is an “AI-generated content” disclaimer enough to comply? It satisfies the obligation, but it is the default option — the one that tells readers nobody checked. For health content, the editorial exception is almost always preferable to the label.
Who is the “deployer”: the agency or the brand? The deployer is whoever uses the AI system under its own authority. In an agency-advertiser relationship, that is settled contractually — but editorial responsibility for the publication stays with whoever publishes.
Is our internal content (intranet, sales decks) concerned? Paragraph 4 covers texts “published with the purpose of informing the public”. Internal and non-public B2B sales materials fall outside this obligation — your web content does not.
Sources
- Regulation (EU) 2024/1689 of 13 June 2024 (AI Act), OJ L, 12.7.2024 — Article 50 (transparency obligations), Article 99 (penalties).
Publishing AI-assisted health content? The Blueprint diagnostic maps your editorial and AI visibility in 8 minutes — and the strength of your editorial responsibility chain is part of the equation. Take the diagnostic → · To scope the topic in person: 30 minutes with Nicolas.