Our method

Where the news comes from

We continuously monitor 23 RSS feeds: French, British, American and Spanish business and general press, specialist HR outlets, research institutes working on working time, and Google News queries in French, English and Spanish.

A first filter discards anything that is not about the subject, based on keywords. Whatever passes that filter is analysed, then either published or set aside.

We never reproduce an article. Every news item links back to the original publication, on the site of the outlet that wrote it. What you read here is a few sentences meant to help you decide whether the article is worth your time.

What the artificial intelligence does

The model we use is Claude Haiku 4.5, from Anthropic. For each article collected, it produces:

• a two- to three-sentence summary, written separately in French, English and Spanish;
• a relevance score from 0 to 100: how much the article genuinely deals with the four-day week;
• a category (pilot scheme, legal framework, company account…);
• the country of the source, when the site's address is not enough to determine it.

The score governs publication: above 80 out of 100 an article is published automatically; below that it stays a draft and appears nowhere on the site until a human picks it up. That threshold is why an article published elsewhere about the four-day week may not appear here.

What the artificial intelligence does not do

It writes none of the substantive pages. The reference page on the four-day week, the practical guides, the "About" page and the one you are reading are written and reviewed by a human, with sources.

It does not rewrite the facts. Summaries condense, they do not fill gaps: when a figure is absent from the original article, it is absent from the summary.

It does not translate quotations. An account published in English stays in its original language, with its translation flagged as such underneath. Re-translating a quotation and presenting it as the original wording would amount to manufacturing a statement.

The illustrations

The images accompanying our news items are generated by artificial intelligence (seedream-4.0 model, via the POE API). None of them is a photograph.

That choice is first of all a legal one: in France, press imagery is protected twice over — the photographer's copyright and the neighbouring right of press publishers. Taking an outlet's thumbnail to illustrate our link would be reproduction without permission. So we generate our own.

Three rules are written into the generation instructions, with no exception possible:

• never a real person, an identifiable face or a public figure;
• never a logo, a brand or the style of a living artist;
• never text inside the image — an image must not be able to assert anything.

The result is deliberately abstract and editorial: an illustration, not a document.

The forum and moderation

Every message posted on the forum goes through two stages.

The first is a blocklist of forbidden terms, applied without any artificial intelligence: insults and hate speech are blocked immediately, including when they are disguised (spaced-out letters, substituted characters).

The second is an AI judgement on the intent of the message: a harsh word inside a quotation from an article is not abuse aimed at someone, and that distinction is one a word list cannot make.

If the AI service fails or slows down, the message goes to human review. It is neither published silently nor blocked silently: a technical failure never decides in our place.

The company directory

The companies listed are submitted by their employees, their management or our readers. Every submission is reviewed before publication: we check that the company exists and that it genuinely runs some form of four-day week. A pending entry does not appear in the directory.

Our limits

An AI model can get it wrong: a poor summary, a poor score, a poor category. When in doubt, the original article prevails — the link is always there, one click away.

If you spot an inaccurate summary, a company wrongly listed or an unsuitable illustration, tell us: we correct it by hand.

Our independence

This site carries no advertising and publishes no sponsored content.

No company pays to appear in the directory, or to rank higher in it.

Books are listed in the library on the strength of their relevance to the subject: nobody pays to be included.