How we work
You are entitled to know how what you are reading was made. Here is the whole process.
Last updated: 3 September 2026
1. Choosing the topic
Every day, a program collects publications from identified tech sources (specialist media, official vendor blogs, project announcements). It discards anything older than 72 hours and anything already covered in the last seven days, then picks one subject.
2. Writing
The article is written by a language model, from the source articles and our editorial guidelines. The rules imposed on it are strict:
- no claim that does not appear in a source;
- no invented figure, date, version or price;
- no quotation that is not in a source;
- two independent sources for any sensitive announcement — acquisition, critical vulnerability, layoffs, financial figures;
- a rumour is labelled as such in the headline;
- never reuse a source's text: facts are summarised and rewritten, with a link to the original.
3. Automated checks
Before any human review, an article is automatically rejected if it cites no source, if a URL is invalid, if it falls outside the length range, or if it covers a subject already published.
4. Human approval
No article is published without being read and approved by Sébastien Soulier. A proposed article can be rejected — it is then never put online. If there is no answer, the article is not published: silence means refusal, never consent.
5. Illustrations
Illustrations are generated by artificial intelligence. They are neither photographs nor news documents: they carry no evidential value and depict no real person. We never use press photography.
6. Errors
This process reduces the risk of error; it does not eliminate it. If you spot one, write from the contact page: the article will be corrected and the correction noted at the foot of the page.
7. Independence
No article is sponsored, commissioned or reviewed by a third party. The site carries no advertising and no affiliate links. Should that change, it would be stated here and on the articles concerned.