Do I have flu or Covid? Why the constant fatigue? What is causing this chest pain? For more than two decades, typing a medical question into the world’s most popular search engine returned a list of links to websites with answers. Today, searching those same health questions on Google often returns an answer written by artificial intelligence.
Google chief executive Sundar Pichai first outlined the company’s plan to build AI into search in May 2024, at its annual developer conference in Mountain View, California. From that month, U.S. users began seeing AI Overviews — AI-generated summaries placed above the traditional list of links — in what amounted to the biggest change to Google’s core product in a quarter century. Google has reported that the feature reached well over a billion users, expanding across more than 200 countries and dozens of languages through 2025.
With search advertising generating a large share of Google’s roughly $200 billion in annual search revenue, the company has moved quickly to defend that business against AI rivals. Pichai has described AI Overviews as “performing well.” But experts warn the format carries real risks.
AI Overviews use generative AI to produce a snapshot answer on top of search results almost instantly. They may cite sources, but they do not reliably know when a source is wrong. Within weeks of the U.S. launch, users documented factual errors across many topics — including some that circulated widely on social media — prompting Google’s head of search, Liz Reid, to acknowledge in a company blog post that in some cases AI Overviews had misinterpreted language on web pages and presented incorrect information. Health, however, is where the stakes are highest.
What a Guardian investigation found
A Guardian investigation reported alarming examples in health-related summaries. In one, AI Overviews returned masses of numbers for liver blood tests with little context — figures that vary by age, sex, ethnicity and nationality — in a way that could lead seriously ill patients to conclude their results were normal and skip follow-up appointments. In another, the summaries gave “grossly inaccurate” information about women’s cancer tests, which experts warned could cause people to dismiss real symptoms. Reporting also highlighted a summary advising people with pancreatic cancer to avoid high-fat foods — close to the opposite of standard guidance, which can matter for tolerating chemotherapy or surgery.
Google initially defended the feature, saying that the “vast majority” of health AI Overviews provide accurate information, that it invests heavily in quality for health topics, and that summaries link to reputable sources and recommend seeking expert advice. After being shown the Guardian’s findings, the company removed AI Overviews for certain queries about liver function test ranges — though reporting found that slightly different wording could still trigger similar summaries.
Where the answers come from
A separate study by SE Ranking analyzed more than 50,000 German-language health searches to see which sources AI Overviews cited most. The single most-cited domain was YouTube. As the researchers noted, YouTube is not a medical publisher but a general-purpose video platform where board-certified physicians and hospitals sit alongside wellness influencers and creators with no medical training. The same study found that roughly two-thirds of citations came from sources not designed to ensure medical accuracy, while academic research and medical journals accounted for well under 1% of citations.
Why presentation, not just accuracy, is the concern
Experts interviewed for the reporting stress that in medicine it is not only where an answer comes from or how accurate it is, but how it is presented. With a traditional list of links, users can compare and critically evaluate competing sources. An AI Overview collapses that into a single confident summary. “Once exposed to an AI summary, users are much less likely to do further research,” said Nicole Gross, associate professor in business and society at the National College of Ireland — losing the chance to compare information or apply their own judgment.
There are further concerns. AI summaries may not distinguish strong evidence from randomized trials from weaker observational findings, and can drop important caveats. Listing claims side by side can make some appear better founded than they are. And because the summaries evolve, the answer to the same question can change even when the underlying science has not. “It means people are getting different answers depending on how they search, and that’s not good enough,” said Athena Lamnisos, chief executive of the Eve Appeal cancer charity.
Google told the Guardian that the links in AI Overviews are dynamic, changing to surface the most relevant and timely information, and that it uses identified errors to improve its systems and takes action where appropriate. For Gross, the deepest worry is that false or dangerous medical advice “gets translated into the patient’s everyday practices, routines and lives” — where, in health care, it “can become a matter of life and death.”
Limitations and what to watch
Several caveats apply. Much of the evidence comes from investigative reporting and a single-snapshot study of German-language queries, so findings may not generalize to every market or persist as the systems change. Google disputes the characterization that most health summaries are inaccurate, and the feature is being actively revised — errors documented at one point may be fixed later, while new ones appear. The reasonable posture is neither to trust nor dismiss AI health summaries wholesale, but to treat them as a starting point: verify anything consequential against a qualified clinician or an established medical source, and be especially wary of specific numbers, dosages and test ranges, which depend on individual circumstances an AI summary cannot see.
The bottom line
AI Overviews have made health information faster to reach and harder to scrutinize at the same time. The convenience is real, but so is the risk when a confident, source-blending summary replaces the act of comparing sources — particularly for questions where the wrong answer carries a genuine cost.
This article discusses health information and medical decision-making. It is not medical advice; anyone with health concerns should consult a qualified professional.
