Google has decided to withdraw some of its health summaries generated with artificial intelligence of search results, after a journalistic investigation showed that they could provide erroneous and potentially dangerous medical information. The move affects functions known as AI Overviews, the quick answer boxes that appear at the top of Google Search.
The company maintains that these summaries are generally useful and reliable for most usersBut the case was uncovered by the British newspaper The Guardian It has put in focus how Google handles health-related searches, an area that is especially sensitive both in the United Kingdom and in the rest of Europe.
Why Google has removed AI Overviews in healthcare
The controversy erupted when The Guardian found that AI Overviews returned medical analysis figures without the proper contextSpecifically, when searching for "what is the normal range for liver blood tests" or "what is the normal result of a liver function test", Google's AI displayed a long list of numerical values ​​presented as normal, without clarifying that these ranges depend on multiple personal factors.
Hepatology specialists and spokespeople for patient organizations warned that this way of presenting information could be misleading for people with serious liver diseaseA user with symptoms or a history of the condition might see their results within these generic ranges, assume everything is fine, and delay a visit to their family doctor or a specialist—something especially worrying in healthcare systems like Spain's, where the early detection it's key.
One of the most criticized points is that the summaries did not take into account basic variables such as the patient's age, sex, ethnicity, or nationalityIn practice, this involves mixing references from different laboratories and countries, creating an appearance of precision that is not actually clinically supported.
Following the publication of the research, Google proceeded to disable AI Overviews for those specific liver function queriesThe affected searches, according to the analysis itself of The GuardianThey stopped showing the AI ​​box, although in many cases the first organic result became precisely the article that denounced the problem.
The most affected health consultations: liver function tests
The most striking examples were related to the well-known liver function tests, very common in both primary care and European hospitals. These tests encompass a set of analyses (such as ALT, AST, bilirubin, alkaline phosphatase, and GGT, among others) that require an individualized clinical interpretation and they cannot be summarized in a single range valid for everyone.
According to Vanessa Hebditch, director of communications and policy at British Liver TrustThe responses generated by Google's AI offered lists of numbers that didn't even always correspond to the specific test that could have been performed on a patient. Furthermore, it was unclear how someone could have seemingly normal results and still suffer from serious, early-stage liver disease.
Hebditch welcomed Google's removal of those summaries, considering it a necessary step to avoid confusion. However, he insisted that the problem goes far beyond a couple of searches: the worrying thing is that the logic behind the construction of AI health overviews may reproduce the same failure in other medical consultations.
The British newspaper itself verified that, by slightly changing the wording of the question to expressions such as "reference ranges for liver function tests" or "LFT reference range", Automatic summaries kept appearingAt least in the first few hours after the case came to light. Only after further checks was it confirmed that Google had also removed those rich results.
Google's official response and its limitations
In response to the criticism, a Google spokesperson stated that the company does not usually... Comment on specific withdrawals within your search engineInstead of providing details about which specific queries have been modified, the representative simply stated that the company is working on "broad improvements" and applies its internal policies when it detects that AI content lacks the necessary context.
The spokesperson added that an internal team of medical professionals reviewed the searches flagged by The Guardian and, according to his version, in many cases The information was not strictly inaccurate and was supported by high-quality websites.The controversy, therefore, focuses less on whether the data was false and more on how it was presented to users without health training.
This clarification is relevant to the European debate on the regulation of AI in healthcare, as it focuses on the responsibility of the large platforms when contextualizing the contentSimply linking to reputable sources is insufficient if the automated summary can lead to misinterpretations on sensitive topics such as diagnostic tests or clinical parameters.
For now, the company has not made any specific public announcements for Spain or other European countries regarding further changes to the AI ​​Health Overviews. However, decisions regarding the search engine are typically applied globally, so Users in Europe will also see AI functionalities affected. when dealing with queries similar to those detected in the United Kingdom.
Concern from patient organizations and health experts
Beyond the specific case of liver function tests, several organizations have taken the opportunity to warn of a broader problem: the difficulty many people have in accessing clear and reliable medical information on the internet. Sue Farrington, president of the Patient Information ForumHe described the removal of these summaries as "a good first step," but stressed that there are still too many examples of incorrect health content generated by AI.
Therefore, these organizations are asking Google and other large technology companies to commit to prioritize links to sanitary applications and reference health organizationssuch as public health services, university hospitals, or recognized patient associations. They also demand that, in clearly sensitive searches (for example, about cancer, mental health, or test results), AI include visible warnings to encourage users to consult a healthcare professional.
Farrington noted that millions of adults in the UK, Spain, and other European countries already struggle to understand traditional medical information, even from official sources. If we add to this... an AI layer that condenses complex content without nuance, the risk of misunderstandings increases even further.
Similarly, technology experts like Victor Tangermann, senior editor of the specialized media outlet FuturismThey believe that the case clearly illustrates that Generative AI systems are not yet ready to replace professional medical guidanceTangermann points to the need to establish clear limits on where and how these types of summaries should be deployed.
Implications for Europe and for medical searches in Spanish
Although the case came to light through a British media outlet, its consequences directly affect the European Union, including Spain. European regulations on artificial intelligence, promoted through the AI ActIt already includes specific requirements for high-risk systems, including the healthcare sector, which could influence how Google and other platforms structure their search products.
In practice, millions of people in Europe use Google as first point of contact for health questionseven before calling their health center or appointment booking service. This means that any flaw in the information displayed, however subtle, has a potential impact on how users interpret symptoms, tests, or treatments.
For countries with public healthcare systems like Spain's, where screening and early diagnosis campaigns are encouraged, it's concerning that a single search result might discourage someone from seeking medical advice. make an appointment with your family doctor This could directly contradict public health policies. Hence the interest of many professionals in having the search engine clearly indicate that the interpretation of test results should always be done by a medical professional.
In bilingual or multilingual environments, such as some regions of Spain, there is also the added difficulty that automatic translations and generated summaries Keep intact the warnings, clinical nuances, and specific reference ranges that may vary by laboratory or country.
This entire debate reopens a fundamental question: to what extent should a tool like Google Search, now enhanced with generative AI, be involved in the interpretation of complex clinical issues? The episode of the withdrawn AI Overviews in healthcare suggests that, unless an appropriate context and a robust medical supervisionAny attempt to simplify health information with artificial intelligence risks creating a false sense of security where the most prudent course of action remains to consult a professional.
