AI search visibility in healthcare: a 2026 US study of how AI assistants recommend doctors
In this AI search visibility study for healthcare, we asked four AI assistants to recommend a US doctor 155 times. ChatGPT, Perplexity and Gemini named specific providers in almost every answer, but the names changed a lot from one run to the next, and each engine leaned on different sources: Healthgrades, WebMD, Zocdoc, Reddit or Google Maps.
How often does ChatGPT recommend specific doctors?
In our November 2026 US panel, ChatGPT named at least one specific provider in 40 of 40 answers to questions like "who is a good dermatologist in Austin for adult acne". The names were unstable: two runs of the same question shared on average 1% to 9% of the providers they named.
Abstract
Patients increasingly ask AI assistants to suggest a doctor. We measured what four assistants do with that request in the United States. Five patient-style questions (dermatology in Austin, plastic surgery in Miami, dental implants in Denver, chiropractic in Phoenix, orthopedic surgery in Chicago) were each asked 8 times per engine on November 19, 2026, giving 155 answers from ChatGPT, Perplexity, Gemini and Google AI Mode. We recorded whether each answer named specific providers, which domains it cited, and how much the named providers overlapped between runs. Three findings stand out. First, chat assistants almost always name providers (98% to 100% of answers), while Google AI Mode did so in 60%. Second, the sources differ sharply by engine: Perplexity cited a doctor directory or review platform in 80% of answers (WebMD doctor profiles most often), ChatGPT in 65% (Healthgrades most often), Gemini in 43% (Reddit and RealSelf), and AI Mode in 23%, relying instead on Google Maps listings in 32 of 35 answers. Third, the recommendations are unstable: two runs of the same question shared on average 1% to 9% of named providers on ChatGPT and 18% to 54% on Perplexity. A practice that checks "does ChatGPT recommend me?" once is reading noise.
Why measure how AI assistants recommend doctors?
Because the question is already being asked. A KFF poll in August 2024 found that about one in six US adults used AI chatbots at least once a month for health information and advice. When the question is "which doctor should I see", the assistant has to choose names, and those names come from somewhere. Practices, and the agencies that advise them, mostly guess at where. We wanted a small, repeatable, published measurement instead of screenshots. We are a marketing agency for physicians based in Israel, so we have an interest in this topic; that is why the method, the counts and a data file are public, and why the findings are reported per engine with intervals rather than as one headline number.
How was the study run?
| Item | Detail |
|---|---|
| Date | November 19, 2026 (single day) |
| Engines | ChatGPT (gpt-5.4-mini with web search), Perplexity (Sonar, US location), Gemini (gemini-3.8-flash with Google Search grounding), Google AI Mode (desktop, US location) |
| Access | DataForSEO APIs, neutral user with no history. API answers approximate the consumer apps; they are not identical to what a logged-in patient sees. |
| Questions | Five patient-style questions, one specialty and one city each (listed below). A sixth question (laser hair removal at a Scottsdale med spa) failed for every engine when our data account ran out of credit, and is excluded. |
| Repeats | 8 runs per question per engine. AI Mode completed 35 of 40 runs (the Chicago question has 3). |
| Answers analyzed | 155 (ChatGPT 40, Perplexity 40, Gemini 40, AI Mode 35) |
| Named provider | An answer that names at least one specific practice or doctor. Names were extracted automatically from bolded text in the answer, so counts of names are approximate. |
| Directory | A known doctor directory or review platform (for example Healthgrades, WebMD doctor profiles, Zocdoc, RealSelf, Castle Connolly, rater8, Yelp, US News). |
| Intervals | Wilson 95% intervals on every percentage |
| Cost | About $2.75 in API fees |
The questions, word for word: "Who is a good dermatologist in Austin, Texas for adult acne?", "Can you recommend a plastic surgeon in Miami for a breast lift consultation?", "I need a dentist in Denver who does dental implants. Who should I see?", "Who is a good chiropractor in Phoenix for lower back pain?" and "Which orthopedic surgeon in Chicago should I see for a torn meniscus?"
How often do AI assistants name a specific doctor?
| Engine | Answers | Named at least one provider | Cited a directory or review platform | Cited a practice or other non-directory site |
|---|---|---|---|---|
| ChatGPT | 40 | 100% (91% to 100%) | 65% (50% to 78%) | 53% (37% to 67%) |
| Perplexity | 40 | 100% (91% to 100%) | 80% (65% to 90%) | 100% (91% to 100%) |
| Gemini | 40 | 98% (87% to 100%) | 43% (29% to 58%) | 100% (91% to 100%) |
| Google AI Mode | 35 | 60% (44% to 74%) | 23% (12% to 39%) | 74% (58% to 86%) |
The three chat assistants treated "who should I see" as a request for names and answered with a shortlist almost every time. Google AI Mode behaved differently: in 40% of its answers it described what to look for and pointed to map results rather than naming providers in the text.
Which sources do AI assistants cite when they recommend doctors?
Each engine has its own habits. The table counts the number of answers (out of 40, or 35 for AI Mode) that cited each platform at least once.
| Engine | Platforms cited most often (answers citing it) | Share of cited domains that were directories |
|---|---|---|
| ChatGPT | Healthgrades 16 · Zocdoc 9 · ASPS (plasticsurgery.org) 6 · Best Pros in Town 6 · RealSelf 5 · hospital physician directories (for example Rush, UChicago Medicine) | 34% |
| Perplexity | WebMD doctor profiles 24 · Zocdoc 16 · rater8 review pages 8 · dentistsranked.com 8 · ASPS 8 · hospital directories 8 | 19% |
| Gemini | Reddit 11 · RealSelf 8 · Castle Connolly 5 · YouTube 5 · Birdeye 2 | 7% |
| Google AI Mode | Google Maps and Business Profile listings 32 · Zocdoc 6 | 9% (plus 36% Google's own listings) |
Three practical readings. Healthgrades and WebMD profiles matter for the two engines that cite directories most, and they are profiles many physicians have never checked. Gemini cited the practices' own websites far more than any directory (86% of its cited domains), plus community sources like Reddit and YouTube. Google AI Mode is, in effect, a layer over Google Maps: the Google Business Profile is the input that counts there. We cover what to fix on each platform in which directories AI assistants cite for doctors.
Do AI assistants recommend the same doctors every time?
No. For each question and engine we compared the provider names in every pair of runs (28 pairs for 8 runs) and averaged the overlap (Jaccard: shared names divided by all names in the two runs).
| Question | ChatGPT | Perplexity | Gemini | AI Mode |
|---|---|---|---|---|
| Dermatologist, Austin | 0.08 | 0.54 | 0.08 | 0.02 |
| Plastic surgeon, Miami | 0.09 | 0.18 | 0.18 | 0.13 |
| Implant dentist, Denver | 0.03 | 0.44 | 0.05 | 0.01 |
| Chiropractor, Phoenix | 0.09 | 0.46 | 0.18 | 0.05 |
| Orthopedic surgeon, Chicago | 0.01 | 0.28 | 0.26 | 0.12 (3 runs) |
Perplexity was the most stable: for three of the five questions, its most-repeated provider appeared in all 8 runs. On ChatGPT, the most-repeated provider for a question appeared in 2 to 6 of the 8 runs, never all 8. This matches what other researchers have reported about list instability in AI answers, and it has a direct consequence: "my practice showed up when I asked ChatGPT" is an anecdote. The useful measure is an answer rate over repeated runs, with an interval. Our AI visibility checker gives you a prompt set and a tally sheet to do that yourself.
What else did the answers mention?
| Signal in the answer text | ChatGPT | Perplexity | Gemini | AI Mode |
|---|---|---|---|---|
| Insurance or in-network | 32 of 40 | 32 of 40 | 19 of 40 | 20 of 35 |
| "Board-certified" | 18 of 40 | 24 of 40 | 25 of 40 | 15 of 35 |
| Reviews | 17 of 40 | 12 of 40 | 14 of 40 | 28 of 35 |
Insurance came up in most ChatGPT and Perplexity answers, often as advice to check networks before booking. Board certification was a frequent qualifier. AI Mode talked about reviews in 80% of its answers, consistent with its reliance on map listings and their star ratings. For a practice, these are the facts an assistant looks for and repeats: accepted insurance, credentials and recent reviews, stated the same way everywhere.
What should a medical practice do with these findings?
- Fix the profiles the engines actually read. Google Business Profile first (AI Mode), then Healthgrades, WebMD and Zocdoc, then your specialty society's find-a-doctor listing. Same name, address, phone, specialties, insurance and hours everywhere.
- Make your own site quotable. Perplexity and Gemini cited practice websites in every answer. A first paragraph that says who you treat, for what, where, and with which credentials is what gets lifted into an answer. See AI SEO for doctors.
- Keep reviews coming, and reply safely. Reviews feed map listings and directory pages. Our guide to Google reviews for doctors covers what Google and the FTC allow and how to reply without disclosing patient information.
- Measure with repeats, per engine. One screenshot tells you nothing given the overlap figures above. Run each question several times on each assistant and count.
None of this guarantees a mention; the engines decide. It improves the inputs they draw on, which is the part a practice controls. If you want help, Only One Medical works with US practices remotely on GEO, SEO and websites, from Israel; our client work so far is in Israel (our numbers).
What are the limitations of this study?
- Small and specific. Five questions, one city each, one day. Results may differ by city, specialty, wording and date. Treat the figures as a snapshot with the intervals shown, not as national rates.
- API, not the apps. We used API access with web search. Consumer apps add personalization, location and conversation history.
- Automated extraction. Provider names were pulled from bolded text. That undercounts names written in plain text and occasionally catches a heading; the overlap figures are approximate, though the gap between engines is large.
- Missing runs. The med spa question and 13 AI Mode runs failed when our data account ran out of credit. We report them as missing rather than estimating them.
- Our interest. We sell GEO services. We have tried to limit that bias by publishing the method, counts and data file, and by naming platforms, not individual doctors.
How can I get the data or cite this study?
The aggregated data (per engine, per question, platform counts, method and limitations) is published as a JSON file: ai-doctor-recommendation-panel-2026-11.json, under CC BY 4.0. We do not publish the raw answer text because it names individual doctors. Researchers can request it by email for verification.
Suggested citation: Fargon, A. (2026). AI search visibility in healthcare: a 2026 US study of how AI assistants recommend doctors. Only One Medical. https://onlyonemedical.co.il/en/insights/ai-search-visibility-benchmark-healthcare-2026
Sources
- KFF Health Misinformation Tracking Poll: Artificial Intelligence and Health Information (August 2024). Share of US adults using AI chatbots for health information.
- Google Search Central: AI features and your website. How AI Overviews and AI Mode select and link sources.
- OpenAI: overview of OpenAI crawlers. The crawlers behind ChatGPT search.
- Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024). The paper that named the field.
Frequently asked questions
Does ChatGPT recommend specific doctors?
In our November 2026 US panel, yes: ChatGPT named at least one specific provider in all 40 answers to five patient-style questions. Perplexity did in 40 of 40, Gemini in 39 of 40 and Google AI Mode in 21 of 35.
Which websites do AI assistants use to recommend doctors?
It depends on the engine. ChatGPT cited Healthgrades most often, Perplexity cited WebMD doctor profiles and Zocdoc, Gemini cited practice websites, Reddit and RealSelf, and Google AI Mode relied on Google Maps listings in 32 of 35 answers.
Are AI doctor recommendations consistent?
Not very. Two runs of the same question on ChatGPT shared on average 1% to 9% of the providers they named. Perplexity was more stable at 18% to 54%. That is why repeated runs are needed to measure visibility.
How was the study done?
Five questions (one specialty and city each) were asked 8 times per engine on November 19, 2026 through DataForSEO APIs with web search, giving 155 answers. Percentages carry Wilson 95% intervals.
Can I use this data?
Yes. The aggregated data file is published under CC BY 4.0. Please cite Only One Medical and link to this page.
Who ran the study?
Only One Medical, a marketing agency for physicians based in Israel, led by Adir Fargon. We sell GEO services, which is why the method and data are public.
Want to know how AI assistants see your practice today?
Send us your website. We will check how ChatGPT, Perplexity, Gemini and Google describe your practice and tell you, in plain English, what is holding you back. Pricing depends on scope, so we quote after the first call.
Keep reading
Fargon, A. (2026). "AI search visibility in healthcare: a 2026 US study of how AI assistants recommend doctors". Only One Medical. https://onlyonemedical.co.il/en/insights/ai-search-visibility-benchmark-healthcare-2026