Google slows down its AI-powered health summaries as pressure mounts for online health security

The way we search Health information on the internet is entering a delicate phaseLarge artificial intelligence models have made the leap from writing emails or summarizing news to offering direct medical answers in the most visible part of search engines, multiplying their impact on patients' daily lives.

In this context, Google has been forced to back down in a key part of their strategy: the AI-generated summaries that appeared at the top of search results for sensitive health queries. After several Due to serious flaws detected by specialists, the company has begun to withdraw these automated responses. in certain searches related to medical tests and diseases.

Google removes AI-generated health summaries after serious errors

In Madrid and the rest of Europe, many users have been living with the so-called AI OverviewsThe automatic summaries that Google generates with generative AI and places above traditional links. The idea is to offer a clear and quick answer, but when it comes to medical questions, the margin of error is minimal.

An investigation by The Guardian It focused precisely on those health summaries. The British newspaper found that, when asked about liver function tests, The system returned lists of supposedly "normal" figures and ranges without clarifying basic aspects such as age, sex, ethnicity, or the patient's clinical context.essential elements for interpreting a blood test.

Hepatology specialists warned that this presentation of data This could lead people with severe liver disease to believe their results were correct. and, consequently, to stop going to medical check-ups or ignore worrying symptoms. It wasn't just an inaccuracy: it was a direct health risk.

Following the publication of these cases, Google decided remove AI Overviews in specific searches such as “what is the normal range for liver blood tests” or “normal liver function test result”. However, the newspaper itself verified that minimal variations in the consultationFor example, changing the wording of the question still generated automatic summaries with the same problem of insufficient context.

The company, true to its communication policy, has preferred not to provide details about each specific removal. A spokesperson simply stated that Google does not comment. “individual eliminations in the search” And that, when AI summaries omit some of the necessary context, they work on general improvements and act according to their internal policies when they deem it appropriate.

Health organizations call for more caution with medical AI

The criticism has not been limited to the technological sphere. Experts in the healthcare sector have called on Google to... greater level of caution when it comes to health issues. Vanessa Hebditch, director of communications and policy at British Liver TrustHe welcomed the removal of some summaries, but warned that the problem remains as long as AI can generate misleading answers by simply slightly modifying the search.

Hebditch noted that liver function tests encompass different types of analysis And interpreting them correctly can be complex even for professionals. An automated list of figures without clarifying which test it is or whether it fits the person's specific case is, in their words, a potential source of confusion which can mask serious pathologies.

Other organizations, such as the Patient Information ForumThey have used this case to highlight a fundamental problem: millions of adults already have difficulty finding reliable medical information And understandable. If search engines incorporate layers of AI that, in practice, compete with traditional sources, the bar for quality should be especially high.

Sue Farrington, president of this forum, described Google's decision to remove some of the summaries as “a good first step”But he insisted that there are still too many examples of automatically generated, incorrect health content. In his view, the search engine's role should be to direct users toward reference health organizations and entities with sound editorial criteria, instead of relying excessively on probabilistic models that do not replace clinical judgment.

Meanwhile, technology and media experts have pointed out that the case highlights an uncomfortable reality: although Google claims that some summaries about cancer or mental health Although they rely on accredited sources and recommend consulting professionals, the system is not free of serious errors. For editors like Victor Tangermann, from the site FuturismWhat happened demonstrates that the deployment of these functions in sensitive areas such as medicine It's not ripe yet..

When AI becomes the gateway to “Dr. Internet”

The case of Google summaries is not an isolated incident. For years, millions of people have used what is known as a "summary service." “Dr. Google” to check symptoms, clarify doubts after a consultation, or prepare before going to the doctor. What's changing now is the technological layer placed between the user and the original sources.

Until recently, the model was relatively clear: The search engine offered a list of links, and each person had to decide which website to trust.Now, AI panels occupy the most visible space and They present the information as a closed and very self-assured answerThis may reinforce the feeling that there is no need to continue searching or comparing.

This shift has direct implications for the Public health in EuropeIn Spain, for example, the regional healthcare systems and the Ministry of Health have invested in expert-validated information portals designed to allow citizens to reliably find answers to their questions. If the search engine's AI layer captures the user's attention before they even reach those websites, The risk of unrigorous shortcuts becoming normalized is evident.

Besides, The way the question is phrased becomes crucialAs British research has shown, changing one expression for another can cause the system to decide whether or not to display an automatic summary. For the average patient, who is unfamiliar with medical jargon and the intricacies of the algorithm, this difference is practically invisible, but can determine whether you receive a prudent response or a ambiguous message.

In a scenario where in-person medical consultations continue to have limited time slots and waiting lists are a concern for a large part of the population, The temptation to delegate more decisions to AI tools is understandableThat's precisely why, The failures in these automated responses are not a minor detail.

Everything suggests that the combination of traditional search engines and generative artificial intelligence will continue to play a central role in how the European population accesses health information. Google's misstep with some medical summaries illustrates the extent to which a misinterpretation can have real consequences and reinforces the idea that, however advanced the models may be, health information needs clinical judgment, context, and human supervision so as not to become an additional risk to patients.