Thousands of AI-driven news sites are gaming Google Discover for profit. French journalist Jean-Marc Manach has tracked over 15,000 such sites, raising urgent questions for publishers and platforms.
AI-generated news sites are rapidly multiplying and siphoning traffic and revenue from legitimate publishers, according to French investigative journalist Jean-Marc Manach. His research has uncovered more than 15,000 such sites, many of which are designed to exploit Google Discover’s recommendation system and monetize through ads.
Manach began noticing unfamiliar news sites in early 2024 through Google Alerts. He found that these sites often featured articles translated or paraphrased from established outlets, or were entirely AI-generated to avoid duplicate content detection. By May 2024, he and his students had identified 75 AI-driven news sites, with at least a quarter containing plagiarized material.
The problem escalated quickly. By September, the list grew to over 250, and with help from fact-checkers at Liberation.fr and Checknews, it surpassed 500 by December. By February 2025, the tally reached 1,000, coinciding with a major legal complaint from more than 40 French media organizations against a GenAI news site accused of plagiarizing up to 6,000 articles per day.
Global tracking firm NewsGuard reported a tenfold increase in AI-enabled fake news sites in 2023. Its team has identified 3,749 AI content farm sites across 16 languages. Manach’s own figures are even higher: 15,000 GenAI news sites in French, 1,500 in English, over 200 in German, and dozens in other languages. While some sites spread disinformation, most are built to generate ad revenue.
SEO professionals are behind the majority of these sites, targeting Google Discover as a primary traffic source. Manach estimates that fewer than 300 editors are responsible for over 75 percent of GenAI news sites, with some individuals managing networks of hundreds or even thousands of domains for SEO and link-building purposes.
Media companies are also adopting GenAI to boost output or replace staff. Manach noted that one major French internet media group announced plans to cut 40 percent of its workforce, and on some of its sites, half the listed journalists were AI-generated personas. He has observed single "journalists" publishing up to 500 articles per day, often with multiple versions of the same story to maximize visibility on Discover or MSN.
Legitimate publishers are increasingly using GenAI to scale content production. Two of France’s largest news editors have launched generative engine optimization (GEO) business lines to sell branded services for mention in large language models and AI chatbots. Last summer, nearly 20 percent of the 1,000 most recommended news sites on Google Discover were AI-generated, as were a third of the top 120 technology news sites on Google News, according to Manach.
Audience data shows the reach of these sites is significant. Mediametrie found that the 250 most recommended GenAI news sites attracted 15-16 million monthly visitors in December-about a quarter of the French population, with 75 percent over age 50. By April, monthly visits rose to 23 million, nearly 40 percent of French citizens.
Financially, the rewards can be substantial. Manach reported that two editors earned over $2 million in just three months from a single English-language site on Discover. Google’s algorithm failed to distinguish between genuine engagement and users scrutinizing deceptive content to file complaints.
Manach warns that the speed and scale of AI-generated news is unprecedented. He has documented cases where a single SEO operator published thousands of articles daily, reaching more readers than major French media outlets. Some articles falsely accused retailers of selling unsafe food, with clickbait headlines generated by AI. He has also seen human journalists inadvertently spread fake news originating from GenAI sites, and local news outlets plagiarized by so-called media startups.
Spotting AI-generated content is becoming more difficult. Manach avoids automated AI detectors due to high error rates and instead relies on forensic and OSINT methods. He examines metadata, author activity, legal mentions, domain ownership, and content structure for signs of automation. He notes that some operators even create fake LinkedIn profiles to legitimize AI-generated journalists.
Unchecked, the proliferation of GenAI news sites threatens to further pollute the information ecosystem. Manach points to a Next.ink user who found that 40 percent of their visited sites were flagged as AI-generated. With regulation and enforcement still uncertain, he urges news organizations to become trusted sources for verified information. Next.ink has released a free browser extension that alerts users when they visit AI-generated news sites.
For publishers seeking to adapt to AI-driven changes in news production, the experience of Jagran New Media integrating Google’s generative AI into its CMS offers a contrasting example of how established outlets are leveraging technology to boost newsroom efficiency and audience growth. Read more about their approach here.
As the AI Act requires clear labeling of AI-generated content, Manach emphasizes that most GenAI news sites remain opaque about their use of automation. Tools and guides are emerging to help journalists and readers detect AI-generated material, but the challenge continues to grow.
