Google pulled its AI answers from some health questions — what that means for your own website’s content

Type a health question into Google and, until recently, you’d often see an AI-generated summary before a single website. In January 2026, that changed for some of those questions — quietly, and only for the ones where Google’s own AI Overviews had been shown to get things wrong. It’s a useful story for any practice that publishes health-related content on its own website, and not for the reason you might first assume.

What Google actually pulled back on

In January 2026, Google removed its AI Overviews from a set of specific health-related search questions, following a Guardian investigation into the accuracy of its AI-generated health answers. The removal wasn’t announced with a blog post or a policy change — it was confirmed only once reporters asked about individual queries the Guardian had flagged (TechCrunch, 11 January 2026; Euronews, 12 January 2026).

The examples that triggered it

• Liver blood-test results. A search for the normal range for liver blood tests returned a wall of numbers with no context for age, sex, ethnicity or nationality — details that change what a “normal” result actually is. Read at face value, it could lead someone to misjudge a genuinely normal result as a problem, or the reverse. Google removed the AI Overview for that question and one close variant, though other phrasings of the same question kept generating a summary (TechCrunch).
• Pancreatic cancer diet advice. A question about diet during pancreatic cancer treatment returned guidance to avoid high-fat foods — advice reviewers said could increase the risk of malnutrition for someone already undergoing treatment (Euronews).
• Other flagged answers. The wider Guardian investigation also raised concerns about summaries on conditions such as psychosis and eating disorders, and one cancer-screening test that was mislabelled outright — and found that asking the same question twice didn’t always produce the same answer (Search Engine Journal, reporting the Guardian’s findings).

Google’s response — narrow, not wholesale

Google didn’t concede the examples were representative of a wider problem. A spokesperson told TechCrunch the company makes “broad improvements” on an ongoing basis, and said an internal review by its own clinicians found that, in many of the flagged cases, the information wasn’t actually inaccurate (TechCrunch). Responding to the Guardian’s original reporting, Google also said some examples were based on incomplete screenshots, and maintained that its AI Overviews link to reputable sources and are on a par with the rest of Search for accuracy (Search Engine Journal).

Whichever account you find more persuasive, what actually happened is clear: the fix was narrow. Google removed the AI Overview from the specific queries named in reporting, not from health search as a whole — and some closely related versions of the same questions kept generating an AI-written answer after the change (Euronews).

How often does this actually come up?

It’s worth knowing how often a health search even produces an AI Overview in the first place, because that decides how many people this touches. An analysis of 146 million search results by the SEO tool Ahrefs found that an AI Overview appeared on 44.1% of medical (“your money or your life”) search queries — more than double the rate across search generally. That figure measures how often an AI summary shows up, not how often it’s wrong; it isn’t an accuracy or error rate, and shouldn’t be read as one (Ahrefs data, reported by Search Engine Journal). But it does mean that when a health search does go wrong, it’s reaching a meaningful share of the people asking.

What this actually means for your own website

None of this is really an argument against AI search, and Google is right that most of its overviews aren’t like the examples above. The genuinely useful takeaway is narrower and more practical: even a system built by one of the best-resourced companies in the world, summarising the entire web, can get a specific, context-dependent health question wrong — because nobody wrote that particular summary for the particular person asking it. Your own website doesn’t have that structural problem, provided what’s on it is accurate, specific, and written by someone who actually knows the subject.

A page on your own site explaining what a result range actually means for your patients, or what to expect around a particular treatment, is doing something a general AI summary can’t do in the same way: speaking to one audience, with facts someone checked, about one situation. That’s not a reason to write more health content carelessly — if anything it’s the opposite. It’s a reason to make sure what’s already on your site is accurate, dated, and reviewed by someone qualified to say so.

Even a system built by one of the best-resourced companies in the world can get a specific health question wrong. Your own page, written for your own patients, is doing a different job.

If you use an AI assistant on your own site

The same principle holds at a smaller scale if your website has its own AI assistant answering patient questions. The risk this episode illustrates — a general-purpose system confidently summarising something it doesn’t have full context for — is exactly what a practice assistant should be built to avoid. A safe one answers only from your practice’s own verified facts, not the open web, and says “I’ll check” when it doesn’t know, rather than generating a plausible-sounding guess. We’ve set out the specific lines a patient assistant should never cross in what an AI patient assistant is — and what it should never do.

Where Tireless fits

We build websites and AI patient assistants that are meant to be trusted with health-adjacent questions, which is why the assistant we ship answers only from a practice’s own checked facts and is tested against real questions — including the ones designed to trip it up — before it goes live. If you want the wider search-visibility picture this story sits inside, our guide to getting found in the age of AI search covers what’s changing for ranking and citations generally. You can see the same standard applied to a real site on our pricing page, or try it directly in the demo below.