Manach works for the independent French website Next.ink and has been a journalist for more than three decades. When Google alerts repeatedly directed him to unfamiliar sites, he took a closer look at them. Their articles were either translated into French from legitimate media outlets, such as Politico or Euronews, or rewritten by artificial intelligence to paraphrase the work of real journalists. The purpose of the rewriting was to avoid accusations of copying content.
By May 2024, Manach had around 75 such websites on his list. Together with his students, he reviewed their content and found that at least a quarter of them were based on stolen articles. But the pace was accelerating. By September, there were more than 250, and he asked fact-checkers from the newspaper Libération and its Checknews section for help.
“In December, the list exceeded 500 AI-generated websites, and in February 2025 it reached 1,000, when we published the first articles,” Manach explains. Publication was delayed for one simple reason. At the time, more than 40 French media outlets filed a joint lawsuit against an artificial intelligence-powered website that copied around 6,000 articles. Every day.
The number is growing every day
Manach has now counted 15,000 artificial intelligence-generated news websites in French alone. To that, he adds around 1,500 in English, more than 200 in German, around 130 in Spanish, and dozens in other languages. He tracks not only those intended to spread disinformation, but primarily those created to generate advertising revenue. According to him, they make up the majority.
A similar picture is presented by the American organization NewsGuard, which has long tracked fraudulent websites. According to the organization, the number of fake-news sites operated with minimal or no human oversight increased tenfold during 2023. So far, the NewsGuard team has found 3,749 content-farm websites in sixteen languages. Czech is among them.
Who creates these websites
Most of the sites were launched by search engine optimization specialists. Initially, they targeted Google Discover because it had become the main source of traffic for news websites. Some were making hundreds or thousands of dollars a day from it.
Manach claims that more than 75 percent of these websites were founded by fewer than 300 people. They are mostly individuals or microbusinesses. Around a hundred of them manage networks comprising more than twenty websites, sometimes more than a hundred, and in extreme cases even more than ten thousand sites linked together to generate backlinks and improve their search rankings.
Media companies themselves are also turning to artificial intelligence. Some use it to “enhance” their journalists, while others replace them outright. “One of the three main French online media groups recently announced that it would lay off 40 percent of its employees. We discovered that on some of its websites, half of the journalists did not exist at all and had been created by artificial intelligence,” Manach says. He also saw a single journalist produce 500 articles a day in several versions of the same story. The aim was to increase the chance that the article would be recommended by Discover or the MSN portal.
Real media outlets also use AI
“Real” media outlets and journalists are also increasingly turning to generated content. It allows them to publish far more articles every day. According to Manach, two of France’s largest news publishers have launched entire business divisions focused on getting their content to appear in responses from language models and chatbots.
The main objective remains recommendations through Google Discover. Last summer, almost 20 percent of the 1,000 most frequently recommended news websites on Discover were created by artificial intelligence. Among the top 120 technology websites in Google News, the figure was as high as one third.
And people are responding to it. Last December, research company Médiamétrie found that the 250 most frequently recommended AI-generated websites were visited by 15 to 16 million people each month, equivalent to a quarter of the entire French population. Three quarters of them were aged fifty or older. By April, this figure had jumped to 23 million people a month, representing almost 40 percent of France’s population.
The profits can be substantial. “Two of these publishers made more than two million dollars from a single website on Discover in just three months, and that was in English,” Manach points out. Google’s recommendation algorithm failed to understand that people spent more time on a deceptive article because they wanted to examine it carefully and write a complaint to the editorial staff—not because they liked it.
When hallucinations cause panic
Manach is sounding the alarm over the speed and scale at which these stories spread. He saw a single optimization specialist publishing thousands of articles a day and reaching more people than all the major French media outlets combined. His articles falsely accused some hypermarkets of selling carcinogenic meat, rotten food washed with bleach, or refrozen products. These alarmist headlines were invented by the artificial intelligence he used.
He has also encountered real, flesh-and-blood journalists who then went on to spread falsehoods invented by these websites. “Artificial intelligence-generated sites spread several different false versions of real events to confuse people. We must acknowledge this and defend ourselves,” he says.
How to spot the fraud
Manach admits that it is becoming increasingly difficult to distinguish AI-generated text and images. He does not trust automated detectors. According to him, they produce too many false results, and some websites even create fake LinkedIn profiles to lend credibility to their fictitious journalists.
He approaches the task like an investigator, focusing primarily on metadata rather than content. He monitors how many articles the same authors publish and over what period of time, and verifies whether the journalist in question even exists. He checks who is listed as the publisher in the masthead and whether it is a real company. He searches the WHOIS database to see how many other domains belong to the same owner or use the same advertising ID. And he looks for articles that are excessively assembled from formatted blocks, FAQ sections, and language that has the feel of machine-generated writing.
If left unchecked, the problem will only continue to grow. Next.ink issued a warning to its readers about artificial intelligence-generated websites, and one user discovered that around 40 percent of the sites they visited were of this kind.
While rules governing artificial intelligence and their enforcement remain largely unresolved, Manach sees a clear role for news media. They should become “safe places” for anyone seeking verified information. Although the European AI Act requires AI-generated content to be labeled, none of these websites disclose their machine authorship. Next.ink has therefore created a free browser extension that alerts users when they enter such a site.
Sources: themediaonline.co. and wan-ifra.org
