Don’t believe everything Google tells you.
Otherwise, you could end up thinking scientists recommend eating rocks, Snoopy assassinated Abraham Lincoln, pizza is better with glue, and a goldendoodle played in the NBA.
These are all false claims presented at the top of Google searches by its artificial intelligence feature AI Overviews.
The tool summarizes results at the top of the page, with many readers taking them as fact and not delving further.
But experts warn that dangerous and less obvious misinformation is being spread with Google’s AI features – making both the company and sources of the misinformation millions of dollars in the process.
‘Whether it’s to boost herbal remedies, supplements or junk, they make money off it,’ Syracuse University Professor of Information Studies Jennifer Stromer-Galley told Daily Mail.
‘We are increasingly not able to trust any information that comes into our search feeds.’
Even after years of improvement by top software engineers, AI results still present wildly outlandish claims.
Google’s AI Overviews feature has come under fire following reports of its misinformation and bizarre claims. Above a Threads user shared that the search engine cited geologists when recommending eating ‘one small rock per day’
Google CEO Sundar Pichai has acknowledged that AI ‘hallucinations’ are an ‘unsolved problem’, but did not give a timeline for resolving it
The latest false story pushed to the top of results, and spotted by Daily Mail this weekend, was the claim that rapper Eminem performed at the funeral of Jeff Bezos’s mother, and that Elon Musk attended. The information was presented before the funeral was even held.
Last month, the Dallas Express spotted Google Overviews claiming without any evidence that Diana Ross was arrested for cocaine possession and went to rehab in 1992.
Google said it gets the ‘vast majority’ of responses right, and its team ‘take action’ when they find mistakes.
Some of the doubtful pronouncements by Google’s AI are ‘hallucinations’, the name for mistakes in its functioning.
But more sinister sites peddling fake news and raking in money from adverts are reportedly gaming Google’s algorithms so that the AI picks them to be prominently featured.
French journalist Jean-Marc Manach has identified over 4,000 sites pushing misinformation that was itself generated by AI, and often promoted by Google.
Manach reported that more than 100 of them are in English, and are tailored to get chosen for Google’s Discover feature, a feed of articles placed prominently in the Google App which many readers use to keep up with current affairs, and which serves up content based on readers’ preferences using AI.
In an interview with Press Gazette last month, he said the suspect list is ‘the tip of the iceberg’, and that uncovering them was a ‘sisyphean task’.
A query about a dog playing in the NBA produced search results affirming a Goldendoodle named Brodie has played in the professional world of basketball
Last month, a news outlet reported that Google’s AI Overviews had produced false claims that singer Diana Ross was arrested for cocaine possession in 1992
False stories from the sites he identified included that banknotes would be removed from circulation in France by October 2025, grandparents will not be able to give money to their grandchildren, and the French government will raid savings accounts directly to finance the war in Ukraine.
Manach said such sites generate their owners ‘thousands of bucks per day in advertising revenues’ by getting huge traffic from people clicking through from Google Discover, then hitting ads – which are run by Google’s Adsense service.
Google takes a cut of advertising revenue through Adsense, meaning traffic to fake news sites is making the giant corporation millions too.
A 2021 University of Michigan School of Information study found 48% of all ad traffic on a sample of 786 ‘fake’ news sites was served by Google.
The researchers concluded Google is the ‘financial engine for fake news’.
False stories on these sites included ‘Scientists Say a Tribe of Apes in West Africa is Close to Developing Their Own Internet’ and ‘McConnell to Co-sponsor Bill Aimed at Restricting Voting Rights to Landowners’, referring to Kentucky Senator Mitch McConnell.
The Verge reported in August 2022 examples of Google’s AI being too eager to erroneously confirm user’s questions.
When asked ‘when did Snoopy assassinate Abraham Lincoln’, Google gave the result ‘1865′, but did not point out that Snoopy is a cartoon character rather than a president-killer.
Some suspicious news stories promoted by Google’s AI-powered features have included Chinese propaganda.
Until November last year, a group of China-linked companies were spawning over 600 websites pushing out thousands of news articles across dozens of countries, laced with ‘narratives aligned to the political interests of the People’s Republic of China,’ Google announced in a report.
The AI tool presented results from fake sources claiming Elon Musk attended the funeral for Jeff Bezos’s mother, AI generated images appeared online showing Elon Musk comforting grieving Bezos – before the ceremony had even taken place
The search results generated on August 21 falsely claimed the funeral took place the day after her passing and reported appearances by Musk and rapper Eminem, ‘who reportedly delivered a moving tribute’
It said the companies worked together in a network called Glassbridge, with web pages that ‘pose as independent news websites’ but instead created propaganda.
Some of those stories made it on to Google’s Discover feature, a feed of articles featured prominently in the Google App which many readers use to keep up with current affairs, and which serves up content based on readers’ preferences using AI.
After discovering the network, Google’s Threat Intelligence Group blocked them from its rankings.
‘We have observed similar behavior from Russian and Iranian [operators],’ said senior threat analyst Vanessa Molter told Forbes in 2024.
The company said it had blocked more than 1,000 sites from Google News and Google Discover since 2022.
Professor Stromer-Galley told Daily Mail: ‘The research on Google Search is that it does influence people’s perceptions and their attitudes.
‘Generally people feel that the information is neutral, and if it’s shown up by Google it is valid.
‘That then exacerbates the problem of people trusting and picking up the information in there, whether it’s in their Google news feed or search results.
Another AI blunder went viral after an X user pointed out that Google suggested adding non-toxic glue to get cheese to stick to pizza, citing a 2013 Reddit post
‘So it doesn’t surprise me at all that there would be efforts to game the AI.’
Political conspiracy theories have also been boosted by Google’s AI Overviews.
Computer scientist Mel Mitchell was dismayed when she saw Overviews falsely claiming Obama was a ‘Muslim president’ in May last year.
‘Given how untrustworthy it is, I think this AI Overviews feature is very irresponsible and should be taken offline,’ she told the Associated Press.
Overviews told another user, Peter Baylies, that there had been only six white presidents, he shared on X.
Google AI Overviews is sometimes confused by even simple trivia or jokes.
A search by Daily Mail on Friday for the ‘biggest island in the Mediterranean’ returned Crete, which is actually in fifth place.
And when satire site The Onion put up a joke story saying a UC Berkeley Geologist recommended eating one small rock a day, Google Overviews presented it to searchers as real.
Computer scientist Mel Mitchell shared her dismay in May last year when she saw Overviews falsely claimed Obama was ‘America’s first Muslim president’
On X, user Peter Baylies shared that Google’s AI Overviews claimed there has only been six white U.S. presidents – curiously all were named James
Google struck a deal last year to mine the online forum Reddit’s data for $60 million annually.
But their AI potentially fell foul of the adage ‘garbage in, garbage out’.
One Reddit user, ‘F**ksmith’, humorously posted 12 years ago recommending putting glue on a pizza to make the cheese stick.
Searchers then spotted last May that Google Overviews was blithely telling people Googling about pizza recipes: ‘You can also add about 1/8 cup of non-toxic glue to the sauce to give it more tackiness.’
A more serious potential health hazard was spotted by Google App users who saw spurious medical news articles presented to them by AI-powered Google Discover, recommending they ‘stop wasting money on insulin’ and take a ‘herbal remedy that Reverses Type 2 Diabetes’.
Professor Stromer-Galley said that studies show readers are simply reading summaries fed to them by AI, not clicking through to the source links and judging their reliability – or providing advertising revenue to sources of real information such as reputable news sites.
A Pew Research center analysis of 68,879 Google searches by 900 people in March 2025 found 15% clicked through on the results when there was no AI summary at the top of the page. Only 8% did when an AI summary was provided.
‘Google is actually hurting its own customers through its AI,’ Stromer-Galley said.
‘I am really quite worried about the changing information environment now with AI-generated search.’
In an interview with Verge that month, Google CEO Sundar Pichai said AI ‘hallucinations’ are an ‘unsolved problem’, and did not give a definite timeline to solve it.
