Google Pulls AI-Written Sites Hosted on Ouest-France and 20 Minutes Domains From Discover

In the night of November 3 to 4, nine sites disappeared from Google Discover, the feed of recommended articles shown on Android phones and in the Google app. What they have in common, according to the outlet Next: articles generated in bulk by artificial intelligence. The surprising detail: four of them were hosted under the web addresses of Ouest-France and 20 Minutes, two major French news outlets.

Next, which has been investigating these sites for more than a year and a half, calls it a “manual penalty,” a sanction decided by Google's teams rather than a simple automated adjustment. The outlet had questioned Google about eight of them on October 30. On November 6, 1492.vision, an observatory that tracks Discover's recommendations, confirmed to Next that these eight sites were among the domains that vanished. The observatory did not supply the list itself; it only confirmed it.

Who are the nine penalized sites?

Three belong to the Reworld Media group: pleinevie.fr, modesettravaux.fr and mariefrance.fr. Five are published by the SICESI group, described by Next as a subcontractor of Ouest-France and 20 Minutes: trucmania.ouest-france.fr, pause-maison.ouest-france.fr, maison.20minutes.fr, journaldesseniors.20minutes.fr and sciencepost.fr. The ninth, lenergeek.com, from the Pool Presse group, appears to have returned to Discover since.

Google's press office did not respond to Next. Its objections are nonetheless known, thanks to an email sent to a publisher and made public by 1492.vision. In it, Google cites “major spam issues,” “scaled content abuse,” cloaking techniques (showing Google different content from what the reader sees) and “repeated or egregious” violations of its policies. To return, publishers must revise their content and then request a review.

How much of your feed do these sites take up?

On their own, these nine sites accounted for about 2% of the articles served by Discover in France, according to 1492.vision, and eight regularly appeared in its ranking of the ten largest AI-generated sites. In October, the observatory counted 169 such sites recommended by Discover. They made up nearly 6% of the articles offered and 7% of recommendations, even though they represent only 2.36% of the 7,173 sites tracked. In mid-October, Next had already shown that a single publisher of this kind, since blacklisted, had published 26,000 articles in one month, compared with 34,000 for Ouest-France.

Google tightened its rules in early October. It now recommends manually reviewing AI-generated content. The search engine values “human work” in its quality assessment and calls fake author profiles a “form of deception.”

How to spot a news site written by AI

Readers don't have to wait for Google's penalties to protect themselves. Next offers a browser extension that flags about 20,000 sites whose articles appear to be largely AI-generated. The outlet does not publish the full list, to prevent it from being used to train AI models. A few habits also help: check that the listed author exists somewhere other than on the site, read the legal notice to find out who really publishes the content, and be wary of a how-to article with no sources or date.

For companies and media outlets, the lesson goes beyond Discover: a subdomain handed over to a third party puts the main brand's name on the line. It remains to be seen whether the penalized sites will return after review, as lenergeek.com seems to have done, and whether Ouest-France and 20 Minutes will keep lending their addresses to this content.