Discovery Science · emerging evidence
Your Patients Are Already Asking ChatGPT Before They Ask You
AI-chatbot use for finding health information roughly doubled in one year, from 16 percent of surveyed consumers in 2024 to 32 percent in 2025, and among people who used AI for health purposes recently, roughly 55 to 59 percent used it specifically to check or research symptoms before seeing a doctor. That is not a future channel a practice can plan for later. It is a present, fast-moving one, and it depends on the same entity-consistency and corroboration work that already decides map-pack inclusion, now doubled onto a second surface most independent practices have never engineered for. This is the evidence, including what it does and does not yet establish for medicine specifically.
A doubling in one year is not a rounding error
Rock Health's 2025 Consumer Adoption of Digital Health Survey found 32 percent of respondents used an AI chatbot to find health information, up from 16 percent in 2024, a doubling in a single year from a credible, established independent research firm. Among people who used AI for health purposes recently, roughly 55 to 59 percent used it specifically to check or research symptoms before seeing a doctor.
Separately, and with a much heavier caveat, OpenAI has reported figures like 40 million weekly ChatGPT health questions and 3 in 5 adults using AI for health in a three-month window. Those numbers come from the AI company itself, a self-interested party with an obvious incentive to report large usage figures, and should be read as directional at best, not independent evidence. The Rock Health finding, produced by a firm with no product to sell into this specific claim, is the stronger basis for the underlying thesis: this behavior is real, and it is growing fast.
The same entity-consistency problem, doubled onto a second surface
The mechanism that determines whether a practice appears in the map pack, an accurate, consistent name, address and phone across the directories a patient or an engine might check, is close to the same mechanism that determines whether an AI engine can confidently name a practice at all. A practice with three slightly different addresses across Google, Healthgrades and Zocdoc, no structured schema on its site, and thin third-party corroboration is a weak input for a map pack. It is an equally weak input for an AI system trying to answer who is actually nearby and legitimate.
That overlap is useful news, not more bad news. A practice does not need a second, separate discipline to address AI-answer visibility. The entity-lock, directory-consistency and schema work that already fixes local search is largely the same foundation an AI-answer engine reads from, engineered once and serving both surfaces.
What makes a source citable inside a generated answer
The peer-reviewed evidence on this question is established, though general-purpose rather than healthcare-specific. The 2024 paper that founded the discipline of Generative Engine Optimization found that adding cited statistics, direct quotations and authoritative sourcing measurably raised a source's visibility inside generated answers, by roughly 30 to 40 percent on average across the systems tested. That such a discipline exists at all is itself a signal: what an AI engine looks for is not the same as what a search-ranking algorithm looks for, and content built to be extracted and quoted performs differently than content built only to rank.
What is also well corroborated, though again from large-sample industry research rather than peer-reviewed studies, is that roughly 40 to 60 percent of AI-cited domains change month over month. An AI-visibility position is not a badge earned once. It has to be actively held, tracked and re-earned as engines change how they summarize local and health options.
What this evidence does, and does not, establish for medicine specifically
The limits matter here more than in most vertical claims, because the underlying behavior is new. No healthcare-specific, primary-sourced study of patient AI-search adoption currently exists at the rigor of, say, the AMA's own physician-practice benchmarking. The Rock Health finding is credible and independent, but it measures general consumer AI use for health information, not a medical-practice-specific study of how patients choose a provider through an AI assistant.
The GEO peer-reviewed evidence and the AI-citation-churn research are similarly general-purpose: they were not run on healthcare queries specifically, and a healthcare-specific GEO study has not been located. The direction is sound, and it is consistent with how this evidence is treated everywhere else on this site, but it should not be read as proof of a healthcare-specific effect size. The behavior is real and doubling, the underlying mechanism for winning it overlaps heavily with local search work a practice should already be doing, and no engine, vendor, or firm can promise a citation.
The evidence
Key findings, with their sources
-
AI-chatbot use for finding health information roughly doubled year over year, from 16% of surveyed consumers in 2024 to 32% in 2025.
emerging Rock Health, 2025 Consumer Adoption of Digital Health Survey, cited in Fierce Healthcare.
-
Among people who used AI for health purposes recently, roughly 55 to 59% used it specifically to check or research symptoms before seeing a doctor.
emerging Rock Health, 2025 Consumer Adoption of Digital Health Survey.
-
Adding cited statistics, direct quotations and authoritative sourcing measurably raised a source's visibility inside generated answers, roughly 30 to 40% on average across the systems tested.
established Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024, arXiv:2311.09735 (peer-reviewed, general-purpose, not healthcare-specific).
-
Roughly 40 to 60% of AI-cited domains change month over month in large-sample industry studies, meaning an AI-visibility position has to be actively held, not won once.
emerging Yext and SISTRIX, large-sample AI-citation studies, 2026 (industry research, not peer-reviewed).
-
OpenAI has reported figures such as 40 million weekly ChatGPT health questions and 3 in 5 adults using AI for health purposes in a three-month window.
contested OpenAI, cited in Fierce Healthcare, 2026 (vendor-sourced, self-interested party, treat as directional only, not independent evidence).
Reference
Glossary
- GEO (Generative Engine Optimization)
- The discipline of structuring content and entity signals so a business is more likely to be cited inside a generated AI answer, distinct from classic search-ranking optimization.
- Entity consistency
- Having identical name, address, phone and identifying details across every directory and profile, so a search or AI engine can confidently resolve them to one business.
- The measured rate at which a business is named across a defined panel of real buyer questions, sampled by engine, locale and date, used in place of an unverifiable promise of a citation.
- AI Overview
- Google's synthesized summary shown above traditional search results for many queries, which can name a small set of local businesses directly inside the summary.
Straight answers
Frequently asked questions
Is AI-search visibility a real channel yet, or is this still hype for a medical practice?
It is real and growing fast, though newer and less settled than classic local search. AI-chatbot use for finding health information roughly doubled year over year, from a credible, independent research firm. The channel is genuinely new, which is exactly why most practices have not yet engineered for it, and exactly why an early, measured read of where a practice stands is worth doing now.
Is there a healthcare-specific study proving this works for medical practices?
No such study currently exists. The strongest peer-reviewed evidence on what makes content citable in AI answers, and the strongest data on AI-citation churn, are both general-purpose, not run on healthcare queries specifically. The direction is sound and consistent with how the rest of this site treats GEO evidence, but it should be read as directional for this vertical, not as a healthcare-specific finding.
How is this different from ranking well on Google?
A page-one Google ranking and an AI-answer citation are two loosely overlapping outcomes, not the same job. Google states plainly that meeting every best practice does not guarantee inclusion in AI Overviews or AI Mode, and there is no markup that forces it. A practice can rank well and still be absent from the AI answer sitting above those results.
Can you guarantee my practice gets cited by ChatGPT or Google AI Overviews?
No. AI-answer selection is undocumented and changes constantly, so a citation cannot be guaranteed. What can be engineered and measured is entity consistency, content structured to be extracted, and third-party corroboration, tracked as a share-of-answer rate across a defined question panel.
Provenance
Sources
- Rock Health, 2025 Consumer Adoption of Digital Health Survey, cited in Fierce Healthcare, 2025 (emerging, credible independent research firm)fiercehealthcare.com
- OpenAI, cited in Fierce Healthcare, 2026 (contested, vendor-sourced, self-interested party)
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024, arXiv:2311.09735 (established, peer-reviewed, general-purpose)arxiv.org
- Yext and SISTRIX, large-sample AI-citation churn studies, 2026 (emerging, industry research)
- Google Search Central, official documentation on AI Overviews and AI Mode inclusion (established, primary source)
Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.