Google has removed some of its artificial intelligence health summaries after a Guardian investigation revealed the technology was providing inaccurate and potentially harmful medical information. The company quietly removed AI Overviews for specific search terms, including “what is the normal range for liver function tests” and “what is the normal range for liver blood tests,” following the report.
The Guardian’s investigation, published in January 2026, found that Google’s AI Overviews, designed to provide quick summaries of search queries, were generating “dangerous and alarming” responses to health-related questions. In one instance, the AI provided misleading information about normal ranges for liver function tests, potentially leading patients with serious liver disease to believe they were healthy. The summaries lacked crucial context, failing to account for factors like nationality, sex, ethnicity, or age.
The issue extends beyond specific test ranges. A subsequent study revealed a broader concern: Google’s AI Overviews are increasingly relying on YouTube as a primary source of information for health queries, surpassing established medical websites. Researchers at SE Ranking analyzed over 50,000 health searches originating in Germany and found that YouTube accounted for 4.43% of all citations within AI Overviews. This is more than 3.5 times the citation rate of netdoktor.de, a leading German consumer health portal, and more than double that of the MSD Manuals, a well-respected medical reference.
The study highlighted that only 34.45% of citations within AI Overviews originated from reliable medical sources. Government health institutions and academic journals comprised a mere 1% of all citations. No hospital network, government health portal, medical association, or academic institution approached YouTube’s level of citation.
Researchers emphasized the inherent risk of relying on YouTube for medical information, noting that the platform hosts content from a wide range of sources, including individuals with no medical training or expertise. One particularly concerning case involved Google’s AI advising patients with pancreatic cancer to avoid high-fat foods – a recommendation directly contradicting established medical guidance and potentially increasing mortality risk. Incorrect information regarding women’s cancer tests was likewise identified, raising concerns that individuals might dismiss genuine symptoms.
The reliance on AI for health information is rapidly growing. OpenAI reports that approximately 40 million people globally are using ChatGPT for healthcare advice daily. A 2026 Health and Media Tracking Survey conducted by the Canadian Medical Association (CMA) found that roughly half of Canadians surveyed consult Google AI summaries and ChatGPT regarding their health. However, the survey also revealed that individuals who relied on this AI-generated advice for self-diagnosis and treatment were five times more likely to experience adverse effects compared to those who did not.
Studies have demonstrated the fallibility of these AI chatbots. Research from the University of Waterloo showed that OpenAI’s GPT-4 provided incorrect answers approximately two-thirds of the time when presented with open-ended health queries. Harvard researchers found that chatbots often failed to challenge nonsensical questions, such as asking for a comparison of the safety of acetaminophen and Tylenol, demonstrating a tendency to prioritize helpfulness over accuracy.
Although acknowledging the convenience of AI-powered health information, particularly for those facing long wait times for specialist appointments or limited access to healthcare, experts caution against blindly accepting AI-generated advice. The potential for misinformation and the lack of critical evaluation remain significant concerns.