Why ChatGPT favors hospitals over independent practices

A patient looking for an orthopedic surgeon in Chicago asks ChatGPT for recommendations. Three names come back, each attached to a hospital system. Five years ago, she would have run that search on Google and sifted through the links herself. Today, artificial intelligence (AI) does the sifting and hands her a short list. Like 84% of patients, she may still read online reviews before booking, but she is choosing among the practices AI mentioned. Practices that ChatGPT leaves out are never considered.

We wanted to know which practices ChatGPT cites. We pulled 200 independent practices at random from the NPPES national provider registry, across 18 specialties and 20 metro areas, and ran approximately 4,950 patient-style questions. ChatGPT cited none of the practices and didn’t mention any of them, either. We expected the number to be low. Zero still surprised us.

A quick word on terminology: a “citation” is the source ChatGPT pulls from to build an answer, and a “mention” is the practice name a patient actually sees. Citations drive mentions because ChatGPT can only name a practice it has a source for. Mentions are what win patients.

AI has reshaped how patients choose doctors, and being left out of AI answers costs practices new patients. Assistants handle roughly a third of US search volume, up about 300% in a year, according to Graphite, and one in three adults in the U.S. has asked an AI chatbot for health information, according to KFF. In Google’s AI Overviews, a brand mentioned in the answer draws about one-third more organic clicks than one that is not mentioned, according to Seer Interactive. For independent practice owners, this is another current pulling patients toward health systems.

Why AI favors hospitals

AI builds an answer from whatever it can cite, so the question is which sources exist. We mapped where ChatGPT’s citations came from:

The numbers are stacked against the average independent practice. The largest single source, hospital staff rosters at 27.5%, is one independents have no equivalent for. Employed physicians land on these pages automatically, and the system’s web team keeps them current. The 23.0% going to practice-owned websites is captured mostly by a handful of large groups and content-rich practices, not the typical independent. The 200 practices we sampled at random earned none of it. Specialty-association and credentialing pages, at 22.1%, are open to any board-certified physician, yet most practices never check what theirs say. The national directories many practices pay for rank only fourth, at 8.3%.

The real gap shows up in which practices get mentioned. Counting hospital-system pages together with the hospital-affiliated physicians listed on them, hospitals take the majority of names on nearly every kind of patient question. Independents, whether practice groups or solo physicians, trail on all of them. The gap runs widest on reputation and discovery questions and closes only on procedure questions, where a practice’s own pages can compete.

These are two different lenses: citations show where ChatGPT gets its evidence, while mentions show who the patient actually sees. Because hospital-affiliated physicians can be supported by several source types at once, their share of mentions can outrun any single citation bucket.

Your specialty and metro decide how often AI mentions your practice

Your odds depend on your specialty and market, not the national average. Cardiology and orthopedics tilt hardest toward hospitals: Hospital-affiliated physicians hold about a third of the cited surface, while independents fall to single digits. Dermatology and plastic surgery tilt the other way. Independents hold most of the cited surface there, and a practice with strong pages can compete.

Geography widens the gap. In markets dense with academic medical centers, independents have less room to surface, so their share ranges from about 7% in Boston to more than 20% in Charlotte, N.C. A practice at 7% may be outperforming in Boston and underperforming in Charlotte. Benchmark against your own specialty and metro, not the national headline.

A compounding edge for the largest systems

The bigger picture follows from the mention data. As more patients start with AI, and AI more often surfaces hospital-affiliated physicians, the largest systems gain a quiet edge in patient acquisition on top of the payer leverage, capital and brand reach they already hold. Independent practices face a new front door that often does not show their name.

This is a snapshot, not a forecast, and AI remains one channel among several. But the direction is familiar: another force making it harder for independents to stay visible without selling to or joining a system.

How to check whether AI mentions your practice

AI mentions the practices whose information it can read, extract, and understand. Check where you stand in about two minutes: Open ChatGPT and ask the questions a patient would ask, namely the best practice in your specialty and city, who to see for a condition you treat, where to get a procedure you perform, and who the top physicians in your area are. Count how many times your practice is mentioned. Zero puts you with the 200 practices we sampled.

These questions follow” First, when AI does mention a practice, where does its description come from, and why does it render one practice in specific, compelling detail (named physicians, fellowship training, signature procedures) while a comparable one down the street gets a generic line? Next, what do the consistently mentioned practices do that the rest don’t?

Nathan Woo is the founder of Halcy (halcy.ai), an AI growth platform for independent health care practices. The research described here was conducted by Halcy and is shared so that practice owners can audit their AI presence regardless of any vendor relationship.

Explore this data by specialty and metro here.

Methodology

The random-sample finding draws on 200 independent practices selected at random from the NPPES registry across 18 specialties and 20 metros, each tested with a fixed battery of patient-realistic prompts scoped to its specialty and metro, about 4,950 queries in all, with multiple runs per prompt to absorb model variance. The mention-share and citation breakdowns come from a larger companion corpus of roughly 35,000 queries, 41,000 mentions and 33,000 citation events. All queries ran against GPT-5.5 through the OpenAI API in May 2026. The work is single-platform and US-only; behavior on Perplexity, Gemini and Google’s AI Overviews may differ. The analysis is correlational, and the cohort figures carry survivor bias.