What Actually Happens When a Patient Asks ChatGPT for a Doctor
When a patient comes in with one complaint, you don’t run one test.
Someone reports fatigue and you’re already working several branches at once: thyroid panel, CBC, maybe a sleep history, maybe a depression screen.
The patient asked one question. You ran a fan-out of sub-questions and then pulled the results back into a single answer.
AI search works on almost exactly that pattern.
One question, many searches
When a patient types “best endocrinologist near me” into ChatGPT or Google’s AI Mode, that query never gets matched against a single result the way an old phone-book listing or a plain keyword search would.
The system parses out what the patient’s actually trying to accomplish, then quietly runs a small batch of narrower searches on the patient’s behalf before it writes a word of the answer.
For “best endocrinologist near me,” the background searches might look something like:
- Endocrinologists within a set radius of the patient’s location
- Which of those doctors treat the patient’s likely condition (thyroid, diabetes, PCOS)
- What former patients say in reviews about wait times, bedside manner, and outcomes
- Which practices show up as accepting new patients
- Board certification and credentials for the doctors under consideration
Each of those sub-searches pulls from different pages, and sometimes from parts of a page rather than the whole document. The system then stitches the strongest passages from all of them into one answer, often with the sources cited.
Google itself acknowledges that an AI Mode answer and a plain search-results page for the same words can come out looking nothing alike, which is a direct consequence of all that background branching. Industry writers call it query fan-out.
Why “rank for one keyword” no longer covers the whole game
The old model of visibility was simple. Your practice ranked for a search term or it didn’t, and that was the whole game.
You optimized the “endocrinologist [city]” page, watched its position, called it done.
Fine as far as it went.
Fan-out makes that model incomplete.
Your practice is now competing across every sub-query the system silently generates on the patient’s behalf, and you’ll never see the list.
A practice website with one thin page about diabetes treatment might win the literal keyword and still lose the fan-out entirely, because it never answers the sub-question about whether the doctor treats PCOS, or never surfaces a review that speaks to wait times.
What the system’s really grading is whether your practice’s presence, across your site, your directory listings and your reviews, holds up to a doctor’s worth of scrutiny spread over five or six angles at once.
Someone still has to rank
Ranking still decides what the AI has to work with. Fan-out just multiplies the number of times ranking happens.
Each sub-query the system generates runs against something close to a normal search index, and the passages that end up in the answer come disproportionately from pages already ranking well for that sub-query.
If your page on thyroid symptoms doesn’t rank for “thyroid symptoms in women over 40,” it isn’t in the pool the AI draws from for that sub-question, no matter how good the homepage looks.
Practices sometimes hear “AI search,” assume classic SEO is obsolete and stop paying attention to it. That’s backward.
Fan-out means classic ranking now has to work correctly across a wider set of specific questions instead of one broad one.
What a comprehensive presence actually looks like
Put that in chart-review terms. A practice that fares well under fan-out usually has a few things going for it.
A page per condition, one per department is too coarse. A single “Endocrinology Services” page can’t answer sub-questions about diabetes, thyroid disorders and PCOS with the specificity each one needs.
Separate pages, each written toward what a patient with that condition is actually asking, cover more of the fan-out.
Reviews that mention specifics. A background search on what patients say pulls more useful signal from one review that names a condition or describes a wait time than from five that say “great doctor.”
You can’t script that. You can ask patients a specific question when you request the review.
Directory listings that match your site. If the AI runs a sub-query on credentials or new-patient status, it’s pulling from wherever that information lives, and that’s often a directory profile.
An outdated Healthgrades or Zocdoc listing can undercut a well-built website in a query the practice never sees.
No dead ends. A sub-query about accepting new patients that lands on a “Contact Us” page with a phone number and nothing else gives the system nothing concrete to cite.
A page that says plainly what conditions are treated, by whom, and how to get an appointment gives it something to work with.
Search it the way a patient would
Pick the condition your practice treats most and search it in plain patient language, the phrasing someone uses before they know the medical term for it.
Then work out what would have to be true across your site, your reviews and your directory listings for an AI system to find five or six good answers waiting.
Wherever there’s a gap, that’s the sub-question your practice is currently invisible for, even if the homepage ranks fine.
Questions practices ask about this
How do I find out whether ChatGPT is already recommending my practice?
Run the searches yourself. Open a fresh chat, set your city, and ask six or eight plain-language questions about the conditions you treat. Write down which practices get named and which pages get cited, then repeat each question two or three times, since answers shift between sessions. Where a rival practice gets cited and yours never does, you have found the page to fix first.
How many condition pages does my practice actually need?
Start with the five to eight conditions that make up the bulk of your visit volume rather than every condition you're technically able to treat. Each page needs enough substance to answer what patients ask before booking, what a first visit involves, and how to make an appointment, usually 600 to 1,000 words. One or two new pages a month is a workable pace for a busy practice.
Can I write these pages myself or do I need to hire someone?
You can write them yourself, and pages drafted by whoever sees those patients often read more specifically than outsourced copy. The parts worth handing off are the ones that eat hours, working out which sub-questions patients actually search, correcting directory profiles one at a time, and keeping review requests going every week. A first pass usually takes ten to twenty staff hours spread over a few weeks.
How long before any of this shows up in search or AI answers?
Plan on months rather than weeks. A new page usually gets indexed within days to a few weeks, then needs longer to build the ranking signal that puts it in the pool AI systems pull from. Directory corrections often propagate faster. Reviews move slowest, since they arrive at whatever rate you ask for them. Be skeptical of anyone promising a specific position or date.
Photograph: hoang anh / Pexels