Will ChatGPT Recommend Your Practice? AEO for Physicians

When a patient asks ChatGPT, Gemini or Perplexity to suggest a doctor, the assistant does not look up a ranking. It searches the web, reads what directories, health systems, insurers, review sites and your own website say, and writes a short answer. Nobody can promise your practice will be in that answer. You can make sure the facts it finds about you are right, and you can check every month in about 20 minutes.
For the bigger picture of how patients move from search to maps to reviews, see How Patients Choose a Doctor Online. For your Google profile itself, see our Google Business Profile guide for medical practices.
What actually happens when a patient asks AI for a doctor?
You will usually get one of three kinds of answers.
- A short named list. Ask for "a good family doctor in Mesa who takes my plan" and you may get three to five practices, each with a line of reasons: specialty, location, rating, maybe "accepting new patients."
- Advice with no names. The assistant tells the patient to check their insurer's directory or a health system's find a doctor page. That still sends them to sources you can influence.
- An answer about your practice. "Is Dr. Lee accepting new patients?" This is where wrong answers hurt most, because the patient may never call to check.
OpenAI's help page for ChatGPT search says answers that use search include inline citations, plus a Sources button listing everything cited. For doctor questions those sources can include physician directories, review sites, health system and hospital pages, insurer directories, map listings and your own website.
AI answers repeat their sources. If a directory still shows a doctor who left in June, the assistant can repeat it confidently.
How many patients are asking AI about health?
About a third of U.S. adults, depending on how you ask. A Pew Research Center survey of 3,488 U.S. adults, fielded June 22 to 28, 2026, found that 34% ever use AI chatbots for at least one of eight health and medical reasons. A KFF poll of 2,480 U.S. adults, fielded May 7 to 31, 2026, found that 29% use AI tools or chatbots for health information or advice at least monthly, up from 17% in June 2024.
Neither survey measured how many people use AI to choose a doctor. The closest Pew item: 15% of U.S. adults use chatbots to help decide whether to see a doctor. In a separate March 2026 poll, KFF found that among adults who used AI for health information in the past year, 18% said not having a regular health care provider was a major reason, and some of those people may be looking for a new practice.
How can a practice manager check AI answers in 20 minutes a month?
Use the same questions on the same day each month and write down what you see.
- Minutes 0 to 2: set up. Open a fresh chat in ChatGPT, Gemini, Perplexity and Google AI Mode, in a browser not signed in to practice accounts. Never type any patient's name or details into a prompt.
- Minutes 2 to 12: ask six questions in each tool, or rotate tools if time is tight:
- "Best [specialty] in [city] accepting new patients"
- "[Specialty] near [neighborhood or ZIP] that takes [your most common plan]"
- "Is [practice name] accepting new patients?"
- "Who are the doctors at [practice name] and what are their specialties?"
- "What are the hours and address for [practice name] in [city]?"
- "Is Dr. [name] board certified, and where does [he or she] practice?"
- Minutes 12 to 17: open the sources. For any wrong answer, or one that leaves you out, note which site supplied the fact.
- Minutes 17 to 20: log it and assign one fix with one owner and one due date.
Here is a sample log for a made up three physician family practice:
| Date | Tool | Question | Mentioned? | What was wrong | Source cited | Fix and owner |
|---|---|---|---|---|---|---|
| Dec 1, 2026 | ChatGPT | Family doctor in [city] accepting new patients | No | Two competitors named, we were not | Two physician directories, a health system page | Check our listing on both directories (front office lead) |
| Dec 1, 2026 | Gemini | Is [practice] accepting new patients? | Yes | Said "not accepting" | Old provider page on our site | Update provider page and status line (web contact) |
| Dec 1, 2026 | Perplexity | Doctors at [practice] | Yes | Listed a physician who left in 2025 | Insurer directory | Submit change to the plan (billing lead) |
Put the next three checks on the calendar now: Tuesday, December 1, 2026, Tuesday, January 19, 2027 (just after HealthCare.gov open enrollment closes on January 15, when patients with new plans start searching, as we cover in our open enrollment post) and Tuesday, February 2, 2027. After that, the first Tuesday of every month.
Which fixes do practices often miss?
The gaps are usually in four places that AI tools lean on and that nobody at the practice owns.
Provider pages with real credentials
Give every physician and advanced practice provider their own page in plain text, not a PDF. Show their name and credentials as licensed, specialty and board certification exactly as the board lists it, training, hospital affiliations, locations and days, languages, and new patient status with a date. Have each provider approve their own page. An assistant asked "is Dr. Lee board certified" can only answer well if a page plainly says so, and you want that page to be yours.
The same facts in every insurer directory
Insurer directories can be among the sources AI answers cite for "who takes my plan," and they are often inconsistent. A 2023 research letter in JAMA compared directories from five large national insurers (UnitedHealth, Elevance, Cigna, Aetna and Humana) for 449,282 physicians listed in more than one of them. Only 19.4% had consistent address and specialty information across all their listings. The authors pointed to group practices that report every physician at every location as a main cause.
Part of the fix is on your side. CMS's No Surprises Act provider overview says providers must submit directory information to plans at the start and end of a network agreement, on material changes and when a plan asks. Its NPI fact sheet says changes to NPI information must be reported within 30 days. And KFF reported in October 2025 that the Medicare Plan Finder added a provider search, and in its testing found some providers listed at the same address as both in network and out of network.
Make one person own a provider data sheet: name, NPI, specialty, locations, days, phone and new patient status. When anything changes, it goes to every plan, the NPI record, hospital pages and major directories in the same week. Your credentialing lead should confirm each plan's steps, and your compliance lead or counsel should confirm what the rules require of your practice.
Physician schema.org types
Structured data labels the facts on a page. Schema.org has a Physician type, described as an individual physician or a physician's office, with properties such as medicalSpecialty, hospitalAffiliation, availableService, usNPI and isAcceptingNewPatients. It also has a more specific IndividualPhysician type with a practicesAt property that links a doctor to the practice where they work.
| Page | Schema.org type | Properties worth filling in |
|---|---|---|
| Each location page | Physician (as the office) or MedicalOrganization | Name, address, phone, opening hours, medicalSpecialty, isAcceptingNewPatients |
| Each provider page | IndividualPhysician | Name, medicalSpecialty, practicesAt, hospitalAffiliation, usNPI |
Keep expectations honest. Google's guidance on AI features says there is no special schema.org structured data you need to add to appear in AI Overviews or AI Mode, and that structured data should match the visible text on the page. Treat markup as a clear label on accurate pages, not a shortcut.
FAQ content written from your front desk
Publish plain answers to the ten questions your front desk hears most: plans accepted, how soon new patients are seen, what to bring, parking. Keep clinical statements out unless a clinician approves them. Google's Search Central documentation says FAQ rich results stopped appearing in Google Search on May 7, 2026. That removed a display feature, not the value of the answers, which patients and assistants still read.
What should you do when AI gets your practice wrong?
Fix the source, not the chat. Correcting a chatbot changes one conversation, not the next patient's answer.
- Find the source. Open the citations. If none are shown, search the wrong fact in quotes on Google to see where it lives.
- Correct it where it lives. Your site first, then directories and insurer listings, then health system pages through your affiliation contact.
- Send feedback in the tool with the thumbs down or feedback option. Treat it as a note, not a fix.
- Check your site is readable. OpenAI says its search crawler surfaces websites in ChatGPT search results and recommends allowing it in robots.txt. Ask your web contact to confirm you are not blocking it.
- Log it and recheck next month before deciding the fix did not work.
If an AI answer makes a false clinical or reputational claim about a named physician, take it to your administrator and counsel before anyone responds publicly.
How do reviews fit in without risking patient privacy?
Review sites can be among the sources AI answers cite, and your public replies are part of what they read. Never confirm that a reviewer is a patient, and never mention a visit, condition, treatment, bill or insurance detail.
HHS has enforced this. In 2023 the Office for Civil Rights announced a $30,000 settlement with a New Jersey psychiatric practice that disclosed a patient's protected health information while responding to a negative online review. This is not legal advice. Your privacy officer or counsel should approve your reply wording; our Google Business Profile guide covers reply rules and sample replies.
What should the next 90 days look like?
| When | What to do |
|---|---|
| By December 1, 2026 | Build the provider data sheet. Run the first check. |
| December 2026 | Provider pages with credentials. Corrections to plans and NPI records. |
| January 19, 2027 | Second check, focused on insurance questions. |
| January 2027 | Front desk FAQ answers. Physician and IndividualPhysician markup. |
| February 2, 2027 | Third check. Compare against December, then keep the monthly rhythm through 2027. |
By spring you will have a record of what AI tools say about you and what you changed. That record matters more than any single answer, because the answers keep changing.
Questions we hear
Can we pay or apply to have ChatGPT recommend our practice?
No. No one can promise a place in an ordinary AI answer. You control whether the facts AI tools find about you are correct, consistent and easy to read.
How often should a practice check what AI tools say about it?
Once a month, plus an extra check after any change to providers, locations, hours or accepted plans.
Do we need schema.org Physician markup to show up in AI answers?
Google says no special schema is needed to appear in AI Overviews or AI Mode. Physician and IndividualPhysician markup can still help machines read your provider pages, as long as it matches what the page visibly says.
Are FAQ pages still worth writing now that Google dropped FAQ rich results?
Yes. Google stopped showing FAQ rich results on May 7, 2026, but plain answers still help patients and give AI tools text to read.
What should we do if ChatGPT says we are not accepting new patients when we are?
Open the cited sources, fix the page or listing that says so, send feedback in the tool and check again next month.
Want to know what AI assistants say about your practice right now? See our medical practice marketing services, or ask for a free AI visibility check and we will show you which answers mention you, which sources they cite and what to fix first.


