TL;DR
A Guardian investigation has found Google’s AI Overviews providing dangerous health misinformation to 2 billion monthly users. Examples include wrong cancer dietary advice that could increase mortality risk and misleading liver function test information that could leave seriously ill patients thinking they’re healthy.
A New Form of Medical Authority
Since launching in May 2024, AI Overviews has become central to Google’s search experience, reaching over 200 countries in 40 languages. The feature uses generative AI to provide single-block answers above traditional search results—a format experts say fundamentally changes how people consume health information.
“Users no longer encounter a range of sources that they can compare and critically assess,” explains Hannah van Kolfschooten, a researcher in AI, health and law at the University of Basel. “Instead, they are presented with a single, confident, AI-generated answer that exhibits medical authority.”
The Guardian found AI Overviews wrongly advising pancreatic cancer patients to avoid high-fat foods—the exact opposite of medical guidance, which experts said could increase mortality risk. In another case, the system provided incorrect liver function test ranges that could lead seriously ill patients to miss follow-up appointments.
The Problem of Confidence
Google initially sought to downplay the findings but subsequently removed some AI Overviews for health queries flagged by the investigation. The company maintains AI Overviews are “reliable” and link to reputable sources.
However, health charities remain concerned. “There are still too many examples out there of Google AI Overviews giving people inaccurate health information,” says Sue Farrington, chair of the Patient Information Forum.
The British Liver Trust’s Vanessa Hebditch notes that removing individual problematic responses doesn’t address the systemic issue: “It’s not tackling the bigger issue of AI Overviews for health.”
Looking Forward
As AI Overviews continues expanding globally, the tension between speed and accuracy in health information becomes increasingly consequential. The fundamental question remains whether conversational AI is appropriate for medical queries where context, nuance, and professional guidance are essential.