Use case assessment
Is AI content generation worth building?
Genuinely AI, universally available, and carrying a disclosure obligation most publishers haven't noticed.
Real capability, no differentiation in building it, and an Article 50 transparency duty on published output that has been enforceable since 2 August 2026.
What it usually means
Producing marketing copy, product descriptions, documentation or social content with a model.
Is it genuinely AI?
Yes, and there is nothing to build. This is a subscription, and the useful work is entirely in what you do around it: what it is allowed to write, who reviews it, and what gets published.
The obligation people have missed
Article 50 requires disclosure of AI-generated text published on matters of public interest, and that became enforceable on 2 August 2026.
Much of the coverage said the AI Act was delayed. The delay applied to the high-risk tier. Transparency was untouched.
There is an exception, and it is the useful part: where the content has undergone human review and a natural or legal person holds editorial responsibility for it, the disclosure duty generally does not apply.
So if you already review before publishing, your gap is documentation rather than a new process — being able to show the review happened and name who is accountable. If you do not review before publishing, the exception is not available to you.
What the simpler version looks like
Decide what you will not generate. That list is more valuable than any prompt library: anything making a factual claim about your product, anything with legal effect, anything a customer would quote back to you.
Then a named editor, and a record. Both requirements are satisfied by a process most content teams already have informally.
What it costs to run
Subscription cost is trivial. Review time is not, and it is the cost that decides whether this is a saving. Generated copy that needs heavy editing takes longer than writing it.
Worth measuring honestly for one month: time from prompt to published, against your previous time from brief to published.
Where it sits under the EU AI Act
Not Annex III high-risk. Article 50 applies to published text on matters of public interest, subject to the editorial-responsibility exception. Providers of generative systems also carry Article 50(2) marking obligations, with a transition to 2 December 2026 for systems already on the market before August.
When it is worth it
For drafts, variants, and formats you already publish at volume with review already in place.
Not for anything making a claim you would be held to, and not as a way to publish more. Volume was never the constraint.
Common questions
Do you have to disclose AI-generated content in the EU?
Article 50 requires disclosure of AI-generated text published on matters of public interest, enforceable since 2 August 2026. The exception is where the content has had human review and a natural or legal person holds editorial responsibility for it.
What does the editorial responsibility exception require?
Human review before publication and an identifiable person or organisation accountable for the content. If you already review before publishing, the work is documenting that the review happened and naming who is accountable.
Does AI content generation actually save time?
Only where review is already in place and the format is one you publish at volume. Generated copy that needs heavy editing takes longer than writing it, which is worth measuring honestly for a month rather than assuming.
Scoring 40 AI use cases in a regulated enterprise — separating what a tool can do from what it should be allowed to publish.