The short version
- Patients are asking AI who to see: in rater8’s 2026 survey (opens in a new tab) of 992 US adults, 47% had used AI to research healthcare providers, up from 31% in 2025.
- AI now rivals referrals: among people who looked for a new doctor, 36% named AI tools as a top influence, against 34% for Google search results and 32% for referrals from another doctor.
- Independents can vanish: in a study reported by Medical Economics (opens in a new tab), ChatGPT cited and mentioned none of 200 randomly chosen independent practices across about 4,950 patient-style questions.
- Hospital rosters feed the answers: in the same study, hospital staff rosters were the largest source of ChatGPT’s citations, at 27.5%.
- Wrong details are common: 66% of patients who had used AI to research providers said they had seen incorrect provider information, such as wrong addresses, phone numbers, insurance details or hours.
A note before you read: this article is about how provider organizations are found in AI answers. It is not medical advice, and it does not suggest that AI assistants should replace a clinician’s judgment.
Who chooses a provider today, and what is a new patient worth?
Patients choose more actively than before, often starting with a local search, and each one can bring years of care.
Switching is common. In rater8’s survey, 72% of patients were in the market for a new provider over the past year: 47% chose a new doctor and 25% searched without switching. When they searched, 55% started with some version of “[specialty] near me.” rater8 sells reputation software to medical practices, so treat its numbers as vendor research; the survey was fielded in April 2026, according to Medical Economics (opens in a new tab).
Most providers now sit inside larger organizations. The American Hospital Association (opens in a new tab) counts 6,100 hospitals in the United States. According to Avalere Health’s analysis for the Physicians Advocacy Institute, reported by Medical Economics (opens in a new tab), 82% of practicing doctors were employed by hospitals or corporate entities at the start of 2026, and those entities owned 63.9% of physician practices, against 29.8% in 2018.
A new patient is worth far more than one visit. The best-known measure is old but telling: in a 2019 Merritt Hawkins survey of 62 hospital finance chiefs, reported by Fierce Healthcare (opens in a new tab), physicians generated an average of $2.38 million a year in net revenue for their affiliated hospitals, through admissions, tests, treatments and procedures. For a health system, the patient who picks one of its physicians brings that downstream care too. For an independent practice, the same patient is the business.
How far has AI entered patients’ search for care?
Far enough that a third of US adults use it for health information, and many ask it about providers.
The KFF Tracking Poll on Health Information and Trust (opens in a new tab) found about a third (32%) of adults turning to AI for health information and advice in the past year. Among those users, 41% said a major reason was to look up information before deciding whether to see a provider, and 18% said they did not have a regular provider or could not get an appointment. Health care professionals (80%) and search engines (68%) remain far more common sources, but many search results now carry AI summaries of their own. Our guide on how telehealth companies get chosen by AI covers patients who turn to virtual care instead.
The platforms describe the scale themselves. OpenAI says that, based on its de-identified analysis, over 230 million people globally (opens in a new tab) ask health and wellness questions on ChatGPT every week. That is documented by the platform, not measured by us.
Provider choice is following. In rater8’s survey, AI tools were a top influence for 36% of people who searched for a new doctor, and for 39% of those who actually switched. Asked which part of a Google results page they trust most when researching a provider, 37% chose the AI Overview, ahead of the regular links and the map.
Where is the line between finding a provider and medical advice?
A provider’s job in AI search is to be found and described correctly, not to diagnose anyone.
The platforms draw the same line. OpenAI says its health experience “is not intended for diagnosis or treatment.” Google has already pulled back where answers went wrong: after a Guardian investigation, TechCrunch reported (opens in a new tab) that AI Overviews were removed for queries such as “what is the normal range for liver blood tests.” Drugmakers work under the same line plus FDA rules, as our guide on keeping drug brands accurate in AI answers explains.
The two kinds of question also look different on Google. In our study of when Google shows an AI Overview, healthcare and dental keywords showed one 43.0% of the time, and 60.0% of them showed a local map pack. Healthcare and dental searches with a map showed an AI Overview 13.3% of the time; those without a map, 87.5%. In plain terms, AI summaries mostly answer the “what is this?” questions, while the map still answers “who near me?”
Our inference for provider marketing: keep two kinds of page separate. Patient education should be written or reviewed by clinicians, sourced and dated. Pages meant to win the booking should state facts a patient can act on: who the physicians are, what they treat, where, which insurance plans are accepted, and how to get an appointment. What tends to appear on cited commercial health pages is covered in what commercial health sites cited by ChatGPT have in common.
What do patients ask AI when choosing care?
Questions that join a specialty, a place, an insurance plan and a reputation check. We wrote these example prompts to show the pattern; they are not recorded patient searches.
| Stage | Illustrative prompt |
|---|---|
| Specialty and place | “Best orthopedic surgeon for knee replacement in Columbus” |
| Access | “Pediatric dentist near me accepting new patients on Medicaid” |
| Insurance | “Which dermatologists in Charlotte take Blue Cross?” |
| Condition to specialist | “What kind of doctor treats long-term dizziness, and who is good for it nearby?” |
| Reputation | “Is [hospital] good for heart surgery? What do patients say?” |
| Comparison | “[Health system A] vs [health system B] for having a baby” |
Only the condition question touches medical information, and even there the useful answer for a provider is a referral to the right specialty, not advice. Every other question is about access, fit and trust, which is exactly the information a provider organization controls.
How does an AI answer become a booked appointment, and what does a miss cost?
The answer names the provider, the patient checks reviews and access, then books. A missing or wrong name breaks the chain.
The path is short: an AI answer → the physician or location page → a rating and review check → insurance and availability → a call or online booking. Each step can end it.
- The rating filter. In rater8’s survey, 75% of patients would not book with a provider rated below 4.0 stars, and 55% had canceled or avoided an appointment because of online reviews, up from 40% in 2025.
- Wrong details. Of the 465 respondents who had used AI to research providers, 66% said they had met incorrect provider information, and 60% of AI users said they trusted the summary without checking it. The same report cites a 2024 study of insurer directories in which physician addresses were consistent only 17% to 28% of the time.
- Our own check. When we asked four assistants for the address, phone, website and hours of real local businesses, answers differed from the Google profile for 17.2% of dentists and 21.1% of physiotherapists.
What it costs is not yet measured in dollars. Our inference: a patient who is given another practice’s name, or your old phone number, rarely comes back to look for you, and each lost new patient takes their future visits with them. If assistants already describe your organization wrongly, here is how to fix wrong brand information in AI answers.
Why do AI assistants name some providers and skip others?
Because they can only name providers they find described in sources, and hospital systems have far more of those sources.
Documented by Google. For local results on Google, the company says ranking is mainly based on relevance, distance and popularity (opens in a new tab), and it asks businesses to provide complete and detailed information. That covers the map, not how assistants choose.
Observed in a study. The Medical Economics study, run in May 2026 by a company that sells AI marketing software to independent practices, tested ChatGPT only. Across its larger set of questions, ChatGPT’s citations came from:
| Source | Share of citations |
|---|---|
| Hospital staff rosters | 27.5% |
| Practice-owned websites, mostly large groups | 23.0% |
| Specialty-association and credentialing pages | 22.1% |
| National directories | 8.3% |
Employed physicians appear on hospital rosters automatically, and system web teams keep them current. The pattern varied by specialty and city: cardiology and orthopedics tilted hardest toward hospitals, dermatology and plastic surgery gave independents more room, and independents’ share ranged from about 7% in Boston to more than 20% in Charlotte.
Observed in our studies. When we asked ChatGPT for the best local providers in services that included dentists and physical therapists, businesses with more Google reviews than the local median were 19.5 points more likely to be listed, after adjusting for map rank and other signals.
Our inference. Assistants appear to favor providers whose facts are published in many consistent, credible places: roster pages, association listings, review profiles and the provider’s own pages. Size helps because it produces those pages, not because assistants prefer hospitals for their own sake. The wider pattern is in do AI assistants favor big brands over smaller competitors?
How does GEO work for a provider organization?
Generative engine optimization (GEO) for providers means publishing accurate, complete facts about physicians, locations and services wherever assistants look.
- A full profile for every clinician. Name, specialty, board certifications, training, conditions treated, procedures performed, locations, languages and insurance accepted, on one page per physician. For independents, this is the equivalent of the hospital roster.
- Consistent facts everywhere. The same addresses, phone numbers, hours and plan lists on the website, Google Business Profile, insurer directories and listing sites. Inconsistent sources produce inconsistent answers.
- Association and credentialing pages. Check what specialty societies and credentialing listings say about each physician, and correct them. These made up 22.1% of citations in the Medical Economics study, and most practices never look.
- Reviews and replies. Ask patients for reviews and reply to them. In rater8’s survey, 66% said a provider’s response to reviews influences their trust. Privacy rules limit what a reply can say about any patient, so keep replies general.
- Service-line pages built for access. For each service, state who it is for, who provides it, where, and how to book. Keep clinical education on separate, clinician-reviewed pages.
- Local coverage. News about new physicians, locations and services in local and specialty press places names next to places and conditions.
- Regular checks. Ask assistants the specialty, access and reputation questions above every quarter, and track which physicians and sources appear.
Health system IT buying is a separate market; for vendors selling to hospitals, see our healthcare software article. Vendors selling to independent practices can see how practice software vendors win demos. GEO cannot guarantee that an assistant names any provider; it makes the facts it finds complete, correct and easy to verify.
What is still unknown about AI and patient choice?
Who gets named is starting to be measured; how many appointments AI answers produce is not.
- Vendor research. The patient survey and the independent-practice study both come from companies that sell to practices.
- One assistant, one month. The citation study tested ChatGPT through its developer interface in May 2026; Gemini, Perplexity and Google’s AI features may differ.
- No booking data. We found no public data linking AI visibility to scheduled appointments or patient revenue.
- Old value figures. The physician revenue survey dates from 2019.
- Policies are moving. Google and OpenAI are both changing how they handle health questions, so today’s answers may not hold.
Where should a provider organization start?
Start by asking assistants the specialty, insurance and reputation questions your patients ask, for your top service lines.
Pick the three service lines and markets that matter most for new patients. Ask ChatGPT, Gemini, Perplexity and Google AI Mode who to see, and note which physicians and locations are named, which sources are cited, and whether your addresses, phone numbers and plans are right.
If new-patient growth in a service line depends on being chosen before anyone calls, ask us to check how AI assistants present your physicians and locations. We will test the questions patients ask in your specialties and cities, find the wrong or missing facts, and plan the profile, review and coverage work that helps patients find and book with you. Our generative engine optimization service page describes how that work is run for provider organizations, from finding the physician facts assistants miss to measuring whether the fixes held.
Frequently asked questions
Do patients use ChatGPT to find a doctor?
Increasingly. In one 2026 survey of 992 US adults, 47% had used AI tools to research providers, and 36% of people seeking a new doctor named AI as a top influence.
Why does ChatGPT recommend hospital doctors over independent practices?
One study found hospital staff rosters were its largest citation source, at 27.5%. Hospital physicians have maintained profile pages; many independents have none.
Should a practice write health content to appear in AI answers?
Only clinician-reviewed, sourced content. Pages meant to win bookings should state facts patients act on: physicians, services, locations, insurance and how to book.
Do online reviews affect whether AI recommends a provider?
They appear to. In our local study, businesses with more reviews than the local median were 19.5 points more likely to be listed by ChatGPT.
Sources
- rater8 (2026), 2026 Patient Choice Report (opens in a new tab)
- Medical Economics (2026), Patients now shop for doctors like consumers, and the bar just got higher (opens in a new tab)
- Medical Economics (2026-07-20), Why ChatGPT favors hospitals over independent practices (opens in a new tab)
- Medical Economics (2026-05-13), Physician independence vanishes as corporate medicine swallows up U.S. health care (opens in a new tab)
- KFF (2026-03), KFF Tracking Poll on Health Information and Trust: Use of AI for Health Information and Advice (opens in a new tab)
- OpenAI (2026-01-07), Introducing ChatGPT Health (opens in a new tab)
- TechCrunch (2026-01-11), Google removes AI Overviews for certain medical queries (opens in a new tab)
- American Hospital Association (2026), Fast Facts on U.S. Hospitals (opens in a new tab)
- Fierce Healthcare (2019), Survey: Physicians net $2.4M in revenue for hospitals each year (opens in a new tab)
- Google Business Profile Help, Tips to improve your local ranking on Google (opens in a new tab)
- Underneath (2026), When does Google show an AI Overview? 1,248 US searches
- Underneath (2026), Do AI answers match a business’s Google profile?
- Underneath (2026), Which Google Maps businesses does ChatGPT recommend?