On Sunday Google pulled a set of its AI Overviews health summaries after a Guardian investigation revealed that the feature could mislead patients with inaccurate lab‑test ranges. The probe focused on queries such as “normal range for liver blood tests,” where the AI displayed raw enzyme values without adjusting for age, sex or ethnicity. Experts warned that a person with serious liver disease could see “normal” numbers and skip follow‑up care. Vanessa Hebditch, communications director at the British Liver Trust, said liver function panels are complex and that the AI’s blunt tables give a false sense of reassurance.Google disabled the liver‑test overviews but left other risky answers, including a pancreatic‑cancer recommendation to avoid high‑fat foods, still live.

The company told The Verge that most health overviews are accurate and that clinicians review the content. Internal analysts, however, trace the problem to a design choice: the AI draws its facts from the top‑ranked web pages, which can be skewed by SEO tricks or outright errors. As the feature gains prominence, regulators and clinicians are watching to see whether Google will overhaul the sourcing model or face further scrutiny.