In short
Twice a year, the Observatory measures how newsrooms in France, Belgium, Switzerland and Quebec publicly disclose their use of artificial intelligence. The method is public. Every media outlet observed has a right of reply.
What is at stake
Since 2 August 2026, Article 50 of the AI Act has required disclosure that a text on a matter of public interest was generated or manipulated by AI. The obligation does not apply if the text has undergone human review or editorial control, and if a natural or legal person holds editorial responsibility for it. Transparency about AI has also become a compliance question for newsrooms.
Do newsrooms publish an AI charter? Do they flag AI in their articles? Do they declare human oversight? To our knowledge, no study measures these practices in a consistent and comparable way over time. The Observatory provides this measurement.
What we measure
For each media outlet observed, we record simple, verifiable indicators:
- whether a public AI charter or policy exists, and how easy it is to find;
- whether content carries an AI disclosure, in articles as well as images;
- the form of that disclosure, from a dedicated page to a label on synthetic images;
- the declared scope of use, when it is stated;
- any mention of human approval before publication;
- any reference to the AI Act or its Article 50.
Each finding is recorded with its date and the web address of the page consulted. Another observer can repeat the measurement.
Method
The Observatory is based on a manual review of a purposive sample of French-language media. We observe public content that any reader can see. Newsrooms’ own statements are not used as a source.
The scope covers four territories: France, Belgium, Switzerland and Quebec. The study runs every six months. Each edition publishes the indicator grid, the composition of the sample and the collection date.
Safeguards
The Observatory is run by Verbatim, which sells AI agents for research, analysis and writing, including to newsrooms. Three rules follow from this:
- no name-and-shame rankings;
- a right of reply for every media outlet;
- no invented data.
First edition
The first edition is in preparation. We will not publish any figures before a data collection carried out according to this method. Until then, this page describes the framework of the study.
Contribute or stay informed
Do you publish an AI charter or an AI disclosure? Point us to a page, exercise your right of reply or ask to receive the first edition.