AI Overviews Cite Only 44.4% of Top Results on Health Queries: Northwestern Study
Google’s AI Overviews appeared for 96.4% of 3,611 benchmark health questions in a Northwestern University study, but they drew on different sources than traditional results. Only 44.4% of the top five organic results were also cited by the AI Overview, which overrepresented commercial publishers such as WebMD and Healthline relative to academic and government sources such as Cleveland Clinic and the NHS. AI Overview-cited pages scored lower than traditional results on four of six credibility criteria drawn from JAMA and HONcode, with small effect sizes. After correcting for classifier error, an estimated 56.7% of cited statements were supported by their cited page, and 20.8% of statements were factual claims with no citation. Queries were run through SerpAPI on September 16-18, 2026, with U.S. English locale; this is an arXiv v1 posting.
Health queries are where AI answers and traditional rankings diverge most. Operators who rely on page-one rank to predict visibility may be missing the sources AI Overviews actually cite.
Use it for: Deciding whether to audit AI Overview citations, not just rankings, for the condition and service queries that drive new-patient demand at each location.
Watch out for: This is an arXiv v1 posting with U.S. English queries from one three-day window, built on benchmark questions rather than local “near me” provider searches. Grading used automated classifiers, and the authors report relative comparisons, so treat the percentages as directional.
Try this: Pull your top 20 service and condition queries per market, record whether an AI Overview appears and which domains it cites, and repeat monthly to track citation share. For context, see the Ahrefs data on AI Overviews and branded searches.
While we aim to share useful and relevant resources, we do not guarantee the accuracy of content on this site or any external links. Views and opinions expressed in referenced content do not necessarily reflect those of Healthcare Growth Strategies.
How accurate were Google’s AI Overviews against physician-written rubrics?
AI Overviews earned 38.2% of all available rubric points, with accuracy at 54.2% and completeness at 21.6%. Among overviews whose rubric anticipated a specific inaccurate or potentially harmful claim, 22.2% contained that content. The authors also report that accuracy was unrelated to how well statements were grounded in their cited sources.