A publisher previously blacklisted for flooding Google Discover with AI-generated and misleading content has returned. At the start of October, three of his sites accounted for up to 4% of Discover traffic. Together they published 26,000 articles, compared with 34,000 for Ouest-France, a newsroom of more than 700 journalists.
Updated October 17 at 10:30 a.m. with clarifications from APIG on the difference between “clicks” and “visits.”
Last week, a report found that traffic to major news sites had fallen by nearly 25% over one year. The decline is thought to be linked, at least in part, to changes in Google’s Discover recommendation algorithm.
Discover is the leading traffic source for news publishers. In June 2025, it generated more than 60% of the traffic for a panel of 68 titles from the Alliance of General Information Press (APIG). It represented 72% of Google-referred traffic, ahead of Google Search (21%) and Google News (7%). Google itself accounted for 70% of total traffic, against less than 8% for social networks.
Analytical (suited to an SEO or industry audience)
Following reports that traffic to leading news sites dropped by almost 25% year over year, attention has turned to Google Discover, whose algorithm changes may be a contributing factor. The dependency is substantial: among 68 APIG titles in June 2025, Discover drove over 60% of traffic. Within Google-referred traffic, it held a 72% share, versus 21% for Search and 7% for Google News. Google as a whole supplied 70% of all traffic, while social networks contributed under 8%.
According to 1492.vision, a dashboard for analyzing content recommended by Discover, the share of articles from news sites that Next identified as AI-generated fell from just over 8% in December 2025 to just over 3% in April. The tool does not claim to be exhaustive, but its figures are considered authoritative within the profession. Since April, the share has fluctuated between 4% and 6%.
In August, the ratio moved between 6% and 7%, then surged in September, approaching 12% in early October. It dropped abruptly to below 5% after Google blacklisted a spammer.
A performance graph comparing the ratio of impressions to unique articles shows that AI-generated sites initially underperformed “human” sites. In June, however, they overtook them, and they continued to outperform human-run sites throughout the summer and into October.
In September, nearly 7% of URLs and 8.5% of hits on French-language websites came from 145 AI-generated news sites. Yet these sites made up just over 2.6% of the 5,552 domains recommended to French-language users that month.
By comparison, the 159 GenAI sites recommended by Discover in June accounted for just under 3% of all domain names identified by 1492.vision, but only 3.32% of URLs and 3.69% of hits. In August 2025, when Discover was flooded with spam, 1492.vision counted 357 GenAI sites. They represented 7.59% of recommended domains, over 9% of URLs and over 8% of hits.
This article is part of Next’s ongoing investigation into the proliferation of AI-generated news sites, which has already produced around fifty pieces. Next has also developed a free extension for Chrome and Firefox that warns users when they visit a GenAI site from its manually compiled list. That database held just over 1,000 sites in February 2025 and now exceeds 19,000, for French-language sites alone.