Google Removes AI Health Overviews After “Misleading” Concerns

For months, Google has sold AI Overviews as a faster way to get “the essentials” without clicking around. The problem is that in health searches, “fast” can also mean missing the one detail that matters.

Now, that tension has turned into a rollback.

After the Guardian investigation flagged misleading health summaries, Google removed AI Overviews for two liver-test queries: “what is the normal range for liver blood tests” and “what is the normal range for liver function tests.”

The context gap: how AI hallucinates “normal” results

The liver-test example wasn’t just “a wrong answer.” It was the kind of answer that can feel accurate while still being dangerous, especially in a Google Search generative experience where the summary is treated like a shortcut.

The reports suggested that the AI Overview surfaced a cluster of reference-range numbers with little context, failing to account for variables like sex, age, ethnicity, or nationality. Health experts warned this could create false reassurance, where someone with serious liver disease might think their results look “normal” and skip follow-up care.

Demographics matter for liver tests because “normal” isn’t universal. For instance, what reads as a normal range for an adult male may be flagged differently for a pregnant woman or a child, where reference ranges and clinical interpretation shift. AI summaries often miss those demographic-specific thresholds and the “talk to a clinician” context that typically accompanies them.

What experts warned could go wrong:

Context-free “normal ranges” can mislead people whose labs require clinician interpretation.

Missing variables (age/sex/ethnicity) increases the risk of medical misinformation from a summary.

The real danger is false reassurance, not just “AI hallucinations.”

Google’s official line

A Google spokesperson told the Guardian the company doesn’t comment on individual removals, and that when AI Overviews miss context, it makes broader improvements and takes action under its policies.

Google also said its internal team of clinicians reviewed what was shared and found that, in many cases, the information was not inaccurate and was supported by high-quality websites. That response is revealing: Google is defending accuracy, while critics are arguing the real failure is context.

The close-variant loophole in Google Search

Even after the two removals, the Guardian reported that slight variations like “lft reference range” and “lft test reference range” could still trigger AI Overviews, and Google said it was reviewing those examples.

TechCrunch added another wrinkle: when it tested those variants later that morning, it didn’t see AI Overviews, but Google still offered AI Mode as an option to ask the same query. So the immediate fix may be query-specific, while the underlying capability remains one click away.

Why this matters more than a normal “bad answer” story

Google Search is the default front door to medical information for millions. The StatCounter data shows Google at about 91% global search market share, which is why mistakes at the top of results can scale fast.

Health groups welcomed the removals but warned that turning off one query does not solve the category. The British Liver Trust described the change as good news but said the same misleading summary can reappear if the question is phrased differently.

The Patient Information Forum’s chair called the removal a first step and urged Google to steer people toward robust, researched health information and trusted care pathways.

What comes next

Google has not said it is pausing health AI Overviews broadly, and it continues to frame the feature as reliable and shown only when the system has high confidence.

But this episode draws a hard line for AI in search: in health, “mostly right” can still be harmful if it encourages the wrong decision. As long as close-variant queries and AI Mode exist, the debate shifts from “did Google remove it?” to “did Google actually contain it?”