• Google deletes AI health summaries

• Experts warn of misleading medical information

Google has deleted some AI-generated health summaries after investigations showed they exposed users to risks from inaccurate or misleading information.

The company said the summaries, which use generative AI techniques to provide overviews of basic information on a topic or question, are “useful” and “reliable.”

However, it noted that some summaries appearing at the top of search results provided inaccurate health information that could endanger users’ lives.

Google displayed incorrect information about basic liver function tests, which could lead patients to mistakenly believe they are healthy.

When searching for “What is the normal range for liver blood tests,” many numbers appeared without clear context or consideration of nationality, sex, or age, potentially causing patients with serious conditions to assume they are normal and skip necessary medical follow-ups.

After the investigation, Google removed AI summaries for searches on “What is the normal range for liver blood tests” and “What is the normal range for liver functions.”

Investigations showed that using similar search queries still triggers AI summaries, raising experts’ concerns, because understanding liver test results is complex and requires more than just comparing numbers.

Hebditch added that the summaries do not warn that normal results can appear even in serious liver disease requiring medical follow-up, representing a major risk to users.

Google said it is reviewing the new examples provided by the investigation, emphasizing that AI summaries only appear when confident in the quality of the information and that it continuously reviews summary quality across topics.

Nevertheless, AI summaries still appear on other topics, notably cancer and mental health, which experts described as “completely wrong” and “very dangerous.”

Google clarified that these summaries are linked to reliable sources, and its internal medical team confirmed the accuracy of most information from high-quality sites.

Experts said this is an important first step, but the bigger challenge is ensuring the accuracy of all AI health summaries, especially as millions worldwide struggle to access reliable information.