AI Overviews in Google Search: A Growing Concern for Health Information Accuracy
Google has integrated artificial intelligence into its search engine through a feature called AI Overviews, which provides information summaries above traditional search results. Launched in May 2024 for US users and expanded to over 200 countries and 40 languages by July 2025, approximately 2 billion people utilize AI Overviews monthly.
"The risks posed by AI Overviews for health are structural, not anecdotal." — Hannah van Kolfschooten, University of Basel
Multiple investigations and studies have raised concerns about the accuracy and sourcing of health-related information provided by these AI-generated summaries.
Accuracy Concerns in Health Information
Medical Advice Inaccuracies
An investigation identified instances where Google's AI Overviews presented health-related content described by experts as inaccurate or misleading. Specific cases included:
-
Pancreatic Cancer Advice: One summary advised individuals with pancreatic cancer to avoid high-fat foods. Anna Jewell, Director of Pancreatic Cancer UK, stated this recommendation is medically incorrect and could compromise patients' ability to receive necessary treatments.
-
Liver Function Tests: Information on liver function tests was described as inaccurate. Pamela Healy, Chief Executive of the British Liver Trust, stated that incorrect "normal" ranges could lead individuals with liver disease to misunderstand their health status and delay follow-up care.
-
Women's Cancer Screening: A Pap test was incorrectly listed as a diagnostic tool for vaginal cancer. Athena Lamnisos, Chief Executive of The Eve Appeal, expressed concern that this could deter individuals from seeking appropriate medical evaluation.
-
Mental Health Information: AI Overviews related to psychosis and eating disorders presented information described as potentially misleading. Stephen Buckley, Head of Information at Mind, noted that advice could lead individuals to avoid professional help.
Organizational Responses
Several health organizations voiced concerns:
- Patient Information Forum: Director Sophie Randall noted the potential for inaccurate health information to be prominently displayed, posing a public health risk.
- Marie Curie: Digital Director Stephanie Parker highlighted the potential for inaccurate or out-of-context information to harm individuals during vulnerable times.
- Google's Response: A company spokesperson stated that many examples were "incomplete screenshots." Google maintains that AI Overviews often link to "well-known, reputable sources" and recommend expert advice. The company indicated it invests significantly in AI Overviews quality, particularly for health topics, and that the majority provide accurate information.
Source Reliability and YouTube Citations
Study Findings
A study by SE Ranking analyzed over 50,000 health queries performed in German from Berlin. Researchers found:
- YouTube was identified as the most cited source in AI Overviews, accounting for 4.43% of all citations (20,621 citations out of 465,823 total)
- No hospital network, government health portal, medical association, or academic institution reached comparable citation numbers
- AI Overviews appeared on more than 82% of health searches examined
Researchers expressed concern, noting that YouTube is "a general-purpose video platform" where content can be uploaded by board-certified physicians, hospital channels, wellness influencers, life coaches, and creators with no medical training.
Expert Commentary
Hannah van Kolfschooten, an AI, health, and law researcher at the University of Basel, stated the findings provide "empirical evidence that the risks posed by AI Overviews for health are structural, not anecdotal." She suggested the reliance on YouTube implies that "visibility and popularity, rather than medical reliability, is the central driver for health knowledge."
Google's Response
Google stated that AI Overviews are designed to surface high-quality content from reputable sources, regardless of format. The company noted that various credible health authorities and licensed medical professionals create content on YouTube. Google also suggested the study's findings, based on German-language queries, might not be generalizable to other regions.
Google indicated that of the 25 most cited YouTube videos in the study, 96% were from medical channels. Researchers cautioned these 25 videos constitute less than 1% of all YouTube links cited by AI Overviews in health contexts.
Philip Morris AI Overview Analysis
Summary Content
A Google AI overview summarizing Philip Morris International relied heavily on the company's own website, presenting a promotional tone that included the company's focus on smoke-free products and sustainability. The overview did not mention the company's history of litigation regarding smoking health risks or its marketing of addictive products.
Researcher Criticism
Professor Becky Freeman of the University of Sydney described the overview as misleading, stating it omitted context about the company's past misconduct. The ABC found similar AI overviews for British American Tobacco and James Hardie that also drew from company sources and lacked negative context.
Expert Observation
Ashwin Nagappa, a researcher at QUT, stated that businesses can structure websites to help AI ingest information, and larger companies with high-traffic websites may have an advantage. Kateryna Kasianenko, also of QUT, noted unpublished research suggesting AI overviews may give prominence to commercial sources, though the reason is unclear.
Discontinuation of "What People Suggest" Feature
Google removed its "What People Suggest" AI search feature, which offered crowdsourced health advice. The company confirmed the removal was part of a "broader simplification" of its search page, stating the decision was not related to the quality or safety of the feature.
The feature had been announced in March of the previous year with the stated aim of providing information from individuals with similar medical experiences. Karen DeSalvo, then Google's chief health officer, stated that while users seek expert medical information, they also value insights from others with comparable experiences.
Broader Context and Expert Concerns
Structural Issues
Experts highlighted that AI Overviews present users with a single, AI-generated answer rather than a range of sources for critical assessment. Hannah van Kolfschooten stated this "actively restructures" health information online, creating "unregulated medical authority" when responses are built on non-medical sources.
User Behavior
Nicole Gross, an associate professor, suggested that users are less likely to research further once an AI summary appears, reducing opportunities for critical evaluation.
Additional Concerns
Experts raised concerns about AI Overviews' potential failure to:
- Distinguish between strong and weak evidence (e.g., randomized trials versus observational studies)
- Include crucial caveats
- Maintain consistency as answers fluctuate, even when scientific understanding remains unchanged
Google responded that links within AI Overviews are dynamic and adapt based on relevance and timeliness. The company stated it commits to using errors to improve its systems.
Previous Incidents
Soon after its US launch, users reported inaccuracies across various subjects, including a false claim about a US president's graduation year. Elizabeth Reid, Google's head of search, acknowledged "oddities and errors" resulting from misinterpretation of web page language.
Future Events
Google's next "The Check Up" event is scheduled for Tuesday, where Chief Health Officer Michael Howell and other staff are expected to discuss new AI research, technological innovations, and partnerships addressing global health challenges.