More than 4,000 fake news websites powered by generative AI have been set up to game Google Discover and search, according to award-winning French journalist Jean-Marc Manach.
The sites are largely in French, but there are already at least 100 in English, and this could be the “tip of the iceberg” Manach told Press Gazette.
The creators of the fake news sites, which spew AI-written articles plagiarised from other sites or which are simply made up, are largely search engine optimisation (SEO) experts.
They aim to either earn money by ‘backlinking’ (artificially boosting the rankings of other sites) or by appearing in Google Discover to harvest ad revenue.
Manach is a French investigative journalist who has worked on topics including privacy and surveillance since the late 1990s and has written about the rise of entirely AI-generated news websites on the French website Next.ink.
He was a pioneer of data journalism in France, working with Reporters Without Borders and Wikileaks, and has won multiple awards for reporting on everything from surveillance to migration, winning a Data Journalism Award and a European Press Prize for his report on 23,000 migrants dying trying to seek refuge in Europe.
The AI sites are robbing real journalists and real publishers of advertising revenue, explains Manach, who started investigating this issue after noticing sites he had never seen before at the start of 2024.
At first, he found just 70 sites, but working with his students at journalism colleges by October he had found such 250 sites. He then worked with journalists from French national title Liberation to investigate the issue further.
He said: “I could never have imagined that those 70 gen AI information websites I had discovered one year ago would let me discover more than 4,000 others since then. It’s a Sisyphean task.”
He believes that at least two site owners have become millionaires using fake AI news sites. His full research is published here.
Google appears unable to block fake news sites from Discover
Press Gazette was shown several of the fake news sites under a non-disclosure agreement, but Manach keeps the full list a closely guarded secret to make it easier to track them. Site operators often move on to new URLs when they are exposed.
A key turning point, Manach said, was when he realised that Google Discover (a news feed within the Google app and on Android devices) was “unable to deal with” these AI news sites.
He said: “Dozens of those editors also try to ‘pop’ on Discover, as it’s a cash machine which lets them hope to earn thousands of bucks per day in advertising revenues.”
Site owners earn money not directly from Discover, but from adverts on their sites served after readers click through. The ads themselves are run through Google Adsense.
Google Discover recommends content to users on Android devices and in Google apps, based on their searches and activities in other apps.
For publishers in the UK, Discover has become an increasingly important source of traffic, as referrals from search have declined. Reach revealed in late 2024 that Discover had become its “biggest referrer” of traffic.
Google’s Discover algorithm also recommended completely false news stories generated by AI, Manach says. These included suggestions that banknotes would cease to exist in France from October 2025, that grandparents would no longer be able to transfer money to their grandchildren, and that the Government would remove money from savings accounts to finance the war in Ukraine.
Other, more surreal stories he says have been surfaced on Google Discover include the claim that a pyramid 25,000 years old had been found under a mountain and that a giant predator resembling an enormous dodo had been found underneath the ice in Antarctica.
Many of these had very obviously AI-generated artwork and titles, Manach said.
He said: “Those generative AI websites tend to ‘hallucinate’ and exacerbate polarising and fake facts and news, as people are more willing to click to know more when the titles of those articles are ‘clickbait’ or frightening. Some editors also don’t hesitate to publish blatantly fake news to attract some views.”
Fake news generated by sites has also been picked up and amplified by human journalists on other sites, such as a story that cars over ten years old would soon be subject to an annual technical inspection.
French Youtube videos revealed how to ‘hack’ Google Discover
The rapid growth of fake AI-powered sites in the French market can be explained by the fact that early adopters began to sell Youtube tutorials on “how to hack Google Discover” using generative AI in 2023 and 2024, Manach explains.
The techniques are widely known among SEO professionals, Manach believes. Now “more and more” site editors are entering the business in other languages, he warned.
Manach said: “Some of the French editors of those gen AI sites I’ve uncovered have already explained that the French market is already ‘overcrowded’ that they began, since, to spot other foreign countries.”
Manach has identified 120 companies and editors behind the sites in France, including ten published by a journalist and media trainer.
Other sites include former news sites which once employed journalists, but have replaced them with AI without telling the readers.
Hallmarks of fake sites include authors who do not exist on other sites, and have no Linkedin page, and illustrations and photographs which do not credit photographers or news agencies by name.
Manach said: “It’s becoming more and more difficult to spot deep fakes and synthetic contents generated by AI, but we must fight this pollution, not only because it threatens journalism, but also our democracies, and the right to be informed.
“I hope many more journalists and fact-checkers will begin, worldwide, to deal with it and document the problem, as Google not only promotes this gen AI content, but also began to replace search links by AI Overviews. What scares me is also the fact that more and more of this gen AI content is itself generated from AI-generated content.”
Google says it has robust policies to counter AI-generated websites
Google says that it has robust policies against spammy websites, and in particular against sites which use AI to generate articles, and polices Google Discover strictly.
A spokesman told Press Gazette: “Our spam-fighting systems aggressively fight mass-produced, low-quality content, keeping the vast majority out of Discover and keeping Search 99% spam-free.
“We have clear policies against content created – regardless if it’s produced by humans or AI – for the primary purpose of manipulating Search rankings and we take action against sites as appropriate.”
As well as the guide to spotting generative AI websites, Manach and his collaborators have created a web extension to pinpoint sites largely created by generative AI.
Manach said: “One of its users told us, for instance, that when he was looking for information about software, he had to click through eight links before finding one that pointed to a website which was not labelled as Gen AI in our database…
“Several others use it to take screenshots and post replies to people who shared links to those GenAI sites on social networks. As an investigative journalist, one thing is to uncover hidden patterns and malpractices, another is to figure out that my findings can help.”
Manach believes that journalists need to ensure that their work stands out from work generated by AI.
He said: “Journalists need to develop new ways to show and prove that their work is better than those gen AI articles. The public need to know whether they are reading ‘real articles’ or synthetic gen AI content, and regulators need to label gen AI content, penalise those who lie about that and are not transparent, and defend ‘real’ journalistic work.”
This is the latest in a series of Press Gazette articles covering the Reality Wars, the rise of fake and AI-generated content in the media.
Email pged@pressgazette.co.uk to point out mistakes, provide story tips or send in a letter for publication on our “Letters Page” blog