Google’s AI Overviews, the summaries that appear atop search results for two billion users each month, have been found to provide “very dangerous” and inaccurate mental health advice, according to a Guardian investigation that has spurred the UK’s largest mental health charity into action.
In response, Mind has launched a year-long commission to examine the intersection of artificial intelligence and mental health, the first such initiative globally. It will convene leading doctors, mental health professionals, individuals with lived experience, policymakers, and tech companies to shape a safer digital mental health ecosystem.
Rosie Weatherley, information content manager at Mind, described how the system has fundamentally changed online information-seeking. “Over three decades, Google designed and delivered a search engine where credible and accessible health content could rise to the top,” she said. “AI Overviews replaced that richness with a clinical-sounding summary that gives an illusion of definitiveness.”
This shift, she argued, often ends the information-seeking journey prematurely, leaving users with “a half answer, at best.” In a test conducted by Mind’s team of mental health experts, AI Overviews generated dangerously incorrect information within minutes of searching with typical mental health queries. The results included assurances that starvation is healthy, that mental health problems are caused by chemical imbalances in the brain, that an imagined stalker was real, and that 60% of benefit claims for mental health conditions are malingering. “It should go without saying that none of the above are true,” Weatherley stated.
The Guardian’s investigation uncovered similar patterns of harmful oversimplification across other health topics. For pancreatic cancer patients, AI Overviews advised avoiding high-fat foods—a recommendation experts label as incorrect and potentially harmful, given such patients often require high-calorie diets. In another instance, a search for vaginal cancer symptoms and tests incorrectly identified a Pap test as the detection method.
Experts have criticised AI Overviews for flattening nuanced and sensitive topics into overly simplistic answers, frequently ignoring crucial factors like age, sex, and ethnicity. Charities have warned that such misleading AI content could deter people from seeking necessary medical care and erode trust in online health information.
Google states it invests significantly in the quality of AI Overviews, particularly for health topics, and maintains that the vast majority provide accurate information. Following the investigation, the company has removed AI Overviews for some queries, though issues persist with other medical and mental health summaries.
Further concerns about reliability are raised by a study from SE Ranking, which found Google’s AI Overviews are two to three times more likely to cite YouTube videos than trusted medical sites in response to health queries.
Despite Google’s disclaimer that AI responses may include mistakes and that users should consult professionals for medical advice, a survey indicated a significant portion of Americans find AI-generated health information reliable and useful. This highlights a potential disconnect between perceived reliability and actual accuracy in critical areas like mental health.
Weatherley criticised what she termed Google’s “whack-a-mole” style of problem-solving, where the company reactively retrains or removes AI Overviews when individuals, organisations, or journalists flag issues. “This feels unserious and not scaled to the size and resource of the company profiting from them,” she said.
She noted that while search engines have evolved to make access to the most harmful results, like suicide methods, less immediate, risks remain. “If you search as an unwell person might search, the risk remains that you will be served harmful inaccuracies and half-truths, presented in calm and confident copy as uncontroversial neutral facts,” Weatherley explained. In searches for crisis information, AI Overviews haphazardly collage various contradictory signposts in long lists.
“Perhaps AI has enormous potential to improve lives, but right now, the risks are really worrying,” she concluded. “Google will only protect you from the potential faults of AI Overviews when it thinks you’re in acute distress. People need and deserve access to constructive, empathetic, careful and nuanced information at all times.”
