A patient with a new prescription opens Google, types a question about side effects, and reads an answer. They never scroll to a link. They never reach your brand site. The answer they trusted was written by Google, assembled from your label, third-party summaries, and whatever the model inferred to fill the gaps. For about half of health searches, that is now the default experience, not the edge case.
Two large-sample studies put numbers on it. WebFX analyzed 130,070 health queries and found that AI Overviews appear on 51.6 percent of healthcare searches, the highest rate of any industry they measured. Ahrefs looked across 146 million SERPs and found that 44.1 percent of medical queries trigger an AI Overview, roughly double the all-query baseline of about 21 percent. Different methods, different sample frames, same conclusion: health is where AI Overviews fire hardest, and the answer the patient reads is produced before the click.
For a pharma brand, that single fact reorders the funnel. The top of your funnel is no longer your homepage or your ranked indication page. It is a paragraph Google writes. And because the queries that trigger AI Overviews are the informational, long-tail ones, the trigger zone overlaps almost perfectly with the questions about your drug that you most want answered correctly.
The trigger zone is exactly your weak spot
The WebFX data shows where AI Overviews fire. Triggering rises with query length and informational intent, and falls for branded queries. AI Overviews appeared on 73.9 percent of health searches with seven or more words and 66.9 percent of informational queries, but only 34 percent of brand-based queries and 25.7 percent of brand-plus-location queries.
That distribution is the strategic point. The searches where Google takes over the answer are the long, plain-language, informational ones: "can I take Varigel with my blood pressure medication," "how long before Varigel starts working," "what are the most common Varigel side effects." Those are the questions a patient asks before they know your brand name, and the questions where an inaccurate or incomplete machine answer is a clinical and regulatory problem, not just a marketing one.
The branded queries, where AI Overviews fire least, are the ones you have always controlled best anyway. Someone who already searches your exact brand name plus your site is deep in your funnel. The AI Overview leaves those mostly alone. It is eating the top, the unbranded informational layer, the part you never built a page for because it never converted directly. That layer was always doing quiet work: shaping what a patient believed before they ever heard your name. Now a model does that work, and you are not in the room.
This is a traffic story and a compliance story at the same time
Most teams will read the 51 percent number as a traffic problem. It is. If half of health searches resolve inside an AI Overview, a measurable share of the clicks that used to land on your unbranded informational content will not happen. That shows up in analytics as slow erosion on the pages that used to bring strangers into the funnel. The cause is not a ranking drop a content refresh fixes. The answer moved above the link.
For pharma the traffic story is the smaller half. The larger half: the AI Overview is a new surface where claims about your drug get made, by a system you do not control, to an audience that treats the answer as authoritative. Google treats health as a Your Money or Your Life category, its highest-stakes content class, and the Ahrefs data shows the consequence: 34.3 percent of all YMYL queries trigger an AI Overview, versus 17.2 percent for non-YMYL queries. The most sensitive content gets the most machine-generated answers.
When that machine answer describes an indication your label does not support, omits a contraindication because the source it leaned on summarized efficacy and skipped safety, or reads a benefit off-label, that is not a missed click. That is an unmonitored claim about your product, published at the top of the most-trusted real estate in search, attached to your brand name, that no one on your team has read. SEO teams measure AI Overview exposure as lost traffic. The function that should also be measuring it, against the approved label, does not currently exist on most brand teams.
And Google is only one engine
The 51 percent figure is the visible part of a larger shift, because Google is not the only place a health question now gets answered. EMARKETER, citing OpenAI data, reports that roughly one in four of ChatGPT's 800 million weekly active users submits a healthcare prompt each week, about 200 million people, with more than 5 percent of all ChatGPT messages globally being health-related and seven in ten of those healthcare conversations happening outside clinic hours. Gallup and West Health find the behavior is now mainstream among patients: 25 percent of Americans have used an AI tool or chatbot for health information or advice, and over half of recent users say they research on their own with AI before or after seeing a doctor.
The AI Overview is the on-ramp, not the whole road. A patient who reads Google's answer this morning asks ChatGPT a follow-up this afternoon and gets a second machine-written paragraph about your drug, from a different model, with different gaps. The unit of pharma visibility is no longer the ranked page on one engine. It is the answer, plural, across every surface where someone now asks. The 51 percent is just the clearest, hardest-measured proof.
What to actually do about it
You cannot opt out of AI Overviews, and you cannot directly edit them. What you can do is make your approved answer the easiest correct answer for a model to lift, then watch what the models actually say. Three moves, in order.
1. Map the informational long tail, not your brand terms. The AI Overview fires on the seven-plus-word, unbranded questions. List the real ones your patients and HCPs ask about the drug, the plain-language versions, not your campaign headlines. That list is your new top of funnel, and right now you almost certainly have no content engineered to win it and no measurement of whether you do.
2. Feed the machine the approved sentence, structured. Models prefer sources that are clear, attributable, and consistent. Your approved indication and safety language, expressed in plain, well-marked, machine-legible content, gives Google and ChatGPT a clean source to prefer over an old abstract or a patient forum. The catch is that the sentence you publish has to be the MLR-approved one, which means the inside of your house has to be in order before you feed the outside.
3. Monitor the answer continuously, against the label. A model answer is not a fixed artifact you screenshot once. It regenerates per user, per engine, and shifts when the model updates or a new source appears. The brands that stay safe will be the ones that detect a new off-label drift the week it surfaces in an AI Overview, trace it to a source, and route it to the people who can correct the underlying content. That requires joining the outside signal to the inside system.
That join is the gap Juncture is built for. Inside, Pre-check compares every asset against the approved label before MLR. Outside, Answer Monitor measures how AI Overviews and answer engines describe your brand and flags off-label drift, continuously, across the surfaces your audience uses. The point is that both are measured against the same approved label, so a machine answer that contradicts your indication shows up as a deviation from a known-good source, not as a screenshot someone forwards in alarm. See how it works on the platform, or bring one brand and we will show you its AI answers today.
The 51 percent number is not the problem. It is the receipt. The answer your patients read about your drug has already moved above the link, on the most-trusted surface in search, for the questions that matter most. Clinicians are doing the same thing: see how HCP AI search behavior is changing, and why GEO is the new SEO for pharma. The brands that start watching the answer now will spend next year defending one they can see. The ones that wait will spend it explaining one they never knew was there.
Sources
- WebFX, "AI Overviews in Healthcare: What 130,070 Health Queries Reveal," 2025. webfx.com
- Ahrefs, "What Triggers AI Overviews? We Studied 146 Million SERPs," 2025. ahrefs.com
- EMARKETER, "1 in 4 ChatGPT users submit prompts about healthcare weekly," 2026. emarketer.com
- Gallup / West Health, "Americans Turning to AI to Supplement Healthcare Visits," 2025. news.gallup.com