If your company writes with AI assistance, of all the concerns raised by Article 50 transparency, there's one we hear more than any other, because it touches the most widespread use of AI there is: "Am I going to have to label every text I write with an assistant as 'AI-generated'? Emails? Proposals? The blog?" It's an understandable worry — and the answer, fortunately, is no. But the "no" has structure, and knowing that structure is what separates well-founded reassurance from naive reassurance.
First: the actual scope of the obligation
The obligation to declare AI-generated text doesn't cover all text. Its scope is specific: it binds deployers of a system that generates or manipulates text, and only as regards text that is published with the purpose of informing the public on matters of public interest. The archetype is journalistic or informational content aimed at a general audience.
That leaves out, from the start, the vast majority of the text a company produces with AI assistance: internal and commercial emails, client proposals, working documentation, private reports. None of that is published to inform the public, and so none of it falls within scope. Anyone who tells you that "from August, all AI-made text has to be flagged" is describing a rule that doesn't exist.
Second: the editorial route for what does fall in scope
And what about content that is published to inform — the corporate blog with an outreach vocation, the open sector analysis, the informational note? Here the Regulation provides a specific route designed exactly for the editorial world: the declaration obligation does not apply where the content has undergone a process of human review or editorial control and a natural or legal person assumes editorial responsibility for its publication.
Read that slowly, because it's the central piece: the legislator isn't targeting AI-assisted text; it's targeting text with no one behind it. If a human reviews what's published and someone — a person or a company — answers for it editorially, the scenario the rule wants to prevent (synthetic content circulating with no one accountable) doesn't arise. The chain of human accountability is, in itself, the transparency the rule is after.
For an SME that publishes carefully worked content — where someone drafts with AI support, someone else reviews, and the company signs off — this route isn't a technicality: it's a description of its normal working flow. The difference between being covered and not being covered isn't changing how you work. It's being able to demonstrate that you work this way.
The word that decides everything: evidence
And here we reach the point this blog repeats until it's tiresome, because it's where everything is decided. The editorial route has two requirements — human review and assumed editorial responsibility — and neither is presumed: both must be evidenced. The day someone asks why your AI-assisted content carries no declaration, your answer can't be "we really did review it." It has to be a record.
A minimum documented editorial workflow looks something like this:
- Written policy: what content the company publishes with AI support and under what process.
- Identified review: for each piece published, who reviewed it and when. A row in a register; no more ceremony is needed.
- Assumed responsibility: the company (or the signing author) expressly assumes editorial responsibility for what's published — the article's byline and the site's legal notice do much of the work if they're consistent.
- Retention: the record is kept like any other compliance evidence — because that's exactly what it is.
Four elements, near-zero marginal cost if integrated into the publication flow, and a disproportionate effect: they turn a diffuse exposure into a defensible position backed by paperwork.
The complete content picture
This piece closes the practical content-transparency triptych we've published this week: the AI Content Transparency Code as the framework of measures whose adequacy the Commission assesses, the official icons as the signal for image, video and audio, and the editorial route as the answer for text. Three different instruments, one and the same underlying logic: Article 50 transparency doesn't demand technical heroics — it demands judgement, process and record-keeping. The usual, applied to the new.
A note on consistency, because we practise what we preach: this very article was drafted with AI support, has undergone human review, and its editorial responsibility is assumed by whoever signs it. That's how simple it is to comply with the editorial route. And that's how simple it is to demonstrate it — if the record exists.
Article 50 series → the four paragraphs on one page.
This article is for informational purposes only and does not constitute legal advice.