The contents of health sites referenced by Google are now used by the search engine to produce a response likely to exempt the Internet user from consulting them.
A little less than a year ago, I wondered about how ChatGPT and other conversational artificial intelligences were disrupting the traffic of health sites.
The threat then seemed to come from outside: after having acquired the habit of questioning Google on their symptoms, their treatments or their medical results, patients began to turn to interfaces capable of providing them with an immediate, personalized response and formulated in a language much more intelligible than that of the countless SEO-optimized articles that they had until then learned to browse.
Since July 22, 2026, the date of the launch in France of Google’s AI Previews and AI Mode, this competition has taken root at the very heart of the search engine. It is no longer just a question, for health sites, of seeing part of their audience move to ChatGPT, Perplexity or Claude, but of noting that Google, after having indexed their content, now uses it to produce a response likely to exempt the Internet user from consulting them.
The historical operation of the engine was, however, based on a form of tacit contract: publishers invested time, money and expertise in the production of information, while Google indexed it, prioritized it and sent them in return a portion of the Internet users who were looking for it.
A blow to the historic contract between Google and producers of written content?
This contract was obviously in no way philanthropic and the engine already captured most of the advertising value, but the AI Overviews profoundly change the balance. When a patient asks what the symptoms of an illness are, how to interpret a biological result or what the side effects of a treatment may be, Google is now able to consult several sources, extract the main elements and bring them together into a coherent answer displayed before the traditional results.
From a user perspective, progress is hard to dispute; information becomes more quickly accessible and the time spent opening and then comparing several pages decreases considerably. But this gain in comfort is based on a shift in value for which information producers risk paying the price: the sites continue to provide the raw material, while Google gradually takes over the relationship with the audience.
A first position which no longer guarantees the click
The first available data already make it possible to measure the extent of the phenomenon. The Pew Research Center analyzed nearly 69,000 searches actually carried out by 900 American Internet users: when an AI Overview was displayed, only 8% of users clicked on a traditional result, compared to 15% when no generated summary appeared.
The presence of this response therefore almost halved the probability that an Internet user would visit one of the referenced sites; barely 1% of searches also resulted in a click to one of the directly cited sources, while 26% of sessions were interrupted after reading the preview, compared to 16% in its absence.
The study conducted by Ahrefs on 300,000 keywords reached a comparable conclusion, estimating that the appearance of an AI Overview is associated with a 58% reduction in the click-through rate to the page occupying the first position.
Academic research based on more than 161,000 pairs of Wikipedia articles exposed or not to the functionality estimates its average causal impact at around 15% of lost daily traffic. These figures do not measure the same sites or the same queries, but they converge towards the same observation: a site can maintain its first position while seeing a significant part of its audience evaporate, because the user obtained what he was looking for before even reaching the first link.
A half-hearted denial from Google
Google claims, however, that the overall volume of clicks sent to the web has remained relatively stable and that the visits generated are now more qualified. This defense isn’t necessarily wrong, but it doesn’t really deny the problem: the total number of searches can increase enough to maintain clicks web-wide, while their distribution shifts profoundly across publishers. This development risks being particularly brutal in health, a sector which concentrates almost all the characteristics favoring the appearance of AI Overviews.
Health concentrates all the risk factors
A WebFX analysis of more than 130,000 US medical queries found their presence in 51% of searches studied, a rate that reached nearly 67% for informational queries and 73.9% when they contained at least seven words.
In the corpus monitored by BrightEdge, coverage even exceeded 90% in several clinical specialties, notably reaching 96% in cardiology and 94% in gastroenterology. The definitions of diseases, lists of symptoms, explanations of treatments, side effects and general interpretations of results which for years constituted the basis of health SEO are therefore the first to be threatened.
They remain essential for feeding and verifying the responses generated, but no longer guarantee that the patient will come into contact with the actor who produced them.
However, exposure is not uniform throughout the health journey. Google mainly absorbs searches allowing us to understand a problem, while those requiring action (find a doctor, make an appointment, consult an image, download a document or use a service) remain relatively better protected.
Beyond traffic, towards a shift in medical authority?
A health establishment or a professional benefiting from strong local visibility will therefore not be affected in the same way as a medical media outlet, an online encyclopedia or a laboratory whose strategy is essentially based on awareness-raising content. But reducing this upheaval to a question of traffic would be a mistake, because users no longer necessarily perceive the responses generated as simple documentary summaries: they gradually grant them real medical authority.
A representative survey conducted in Germany among 1,145 people revealed that 45% of respondents had already asked a chatbot about their symptoms or health; among these users, 55% said they trusted him, 50% felt they understood their symptoms better thanks to him than with traditional research, 30% considered his response to be comparable to a second medical opinion and 16% had already preferred his recommendation to that of a doctor.
When a patient struggles to get an appointment, does not understand a report or simply wants to know if their concern warrants a consultation, artificial intelligence occupies a space left vacant by a healthcare system that is difficult to access and by medical information whose complexity has long been underestimated.
AI Overviews now give this answer the appearance and legitimacy of Google: where previously the Internet user had to choose between several sources and assess their credibility, they receive a single summary formulated with enough confidence to appear as a conclusion rather than a hypothesis.
However, an audit of 1,508 queries relating to pregnancy and early childhood noted contradictions between the AI Overview and the featured snippet displayed on the same page in 33% of cases, while medical precautions were only present in 11% of the responses generated. Another study carried out on more than 55,000 queries estimates that 11% of the assertions produced are not really supported by the cited pages and that almost 30% of the domains used do not even appear in the classic first page results.
Produce what Google cannot summarize
The question posed to healthcare stakeholders therefore goes well beyond digital marketing: what happens when Google uses medical expertise produced by others, simplifies the content and becomes the interface through which millions of people understand their health?
It would be tempting to announce that AI Overviews will cause French sites to lose 30, 40 or 50% of their audience, but such a figure, as effective as it may be in attracting attention, would have no general value: the impact will depend on the share of traffic coming from Google, the proportion of informational queries, the frequency of appearance of the summaries and the drop in the click rate on each of the subjects concerned.
The first organizations under threat will probably be those which have built their visibility on an accumulation of generic content, sometimes very well positioned, but offering nothing that Google cannot summarize in a few paragraphs. Conversely, players with a strong reputation, proprietary data, an identifiable service or a direct relationship with patients and professionals should resist better. We must therefore start by measuring something other than positions: frequency of appearance of AI Overviews, presence of the brand among the citations, share of voice compared to competitors, accuracy of the information included and simultaneous evolution of impressions, clicks and named searches.
Being cited, however, will not recreate yesterday’s traffic: according to Seer Interactive, the citation in an AI Overview is associated with 120% more clicks per impression compared to no citation, but remains 38% less efficient than a search without AI Overview. GEO will therefore above all allow you to lose less visibility than its competitors.
Moving from passive content to a push strategy
The real answer is to produce what artificial intelligence cannot simply piece together from ten similar articles: proprietary studies, real-life data, methodologies, testimonials, identifiable expertise, tools, simulators, questionnaires or resources to accomplish the next step in the journey.
If content offers nothing more than what Google has just summarized, the user has no reason to consult it. Healthcare players will also have to rebuild a direct relationship with their audiences thanks to newsletters, communities, professionals, patient associations, video formats and embodied speaking engagements, in order to no longer depend entirely on traffic that they have never truly owned.
AI Overviews will probably sign neither the death of SEO nor that of medical sites, but they put an end to a rule which seemed stable enough to structure the entire sector: producing the best answer no longer ensures that the user will visit the site which produced it.
The future therefore does not belong to those who publish a new definition of the same symptoms with slightly different words, but to those who provide data that no one else has, an expertise that we can identify, a service that we can use and a relationship that the Google summary cannot replace.
