An off-site citation map for AI patient-discovery, built from 33,340 ChatGPT citation events spanning 15,283 URLs, across 10 specialties and the 20 metros we sampled. The vendor pitch ranks national directories first. The data ranks them fourth.
Off-site citation sources differ by specialty and metro. In most markets, the largest lanes are free sources that practices rarely manage, such as hospital staff rosters and specialty-association credentialing pages.
Where the 33,340 citation events actually land. Hospital systems, practice-owned domains, and specialty associations together carry 72.6%. National directories – the standard vendor pitch – are fourth at 8.3%. Pick a specialty to see how the map reshapes.
| # | Source bucket | Cites | Share |
|---|---|---|---|
| 1 | Hospital systems + AMC staff rosters | 9,161 | 27.48% |
| 2 | Practice-owned domains | 7,684 | 23.05% |
| 3 | Specialty associations + credentialing | 7,363 | 22.08% |
| 4 | National medical directories | 2,777 | 8.33% |
| 5 | State medical boards | 1,706 | 5.12% |
| 6 | Review aggregators | 1,275 | 3.82% |
| 7 | Federal authority (NIH, CDC, FDA) | 1,087 | 3.26% |
| 8 | Regional city magazines | 886 | 2.66% |
| 9 | Ranking aggregators (Newsweek, MediFind) | 755 | 2.26% |
| 10 | Lab + imaging chain locators | 432 | 1.30% |
| 11 | Other (Castle Connolly, news, misc) | 214 | 0.64% |
Hospital-system staff rosters and specialty-association credentialing pages together carry 49.6%. National directories carry only 8.3% – and most practices have two or three vendors there.
Get providers added to the staff rosters of the hospital systems you affiliate with before renewing another directory subscription.
Plastic surgery: specialty-association credentialing at 30.1%. Cardiology: hospital systems at 33.4%.
Filter the map above by your specialty before deciding which lanes to invest in. The all-specialties mean is the wrong target for any single cell.
Review aggregators are 3.8% combined, Yelp is 0.11%, and Google Business Profile has zero direct citations.
Reviews and Google Business Profile support patient conversion rather than direct ChatGPT citations, so they should keep their own budget.
By metro
In Tampa, 40.5% of what ChatGPT cites is a practice-owned domain. In New York it is 10.5%. The three buckets that lead nationally carry at least 63% of every market, but the order they fall in – and how much room they leave an independent practice – is a local question.
The regional-magazine pipeline is the cleanest on-or-off split: it carries 9.2% of Seattle, 7.3% of San Diego and 6.5% of Boston, and was cited not once in New York, St. Louis or Baltimore. Open a metro for its full mix.
Tampa is the most practice-owned market in the corpus. Two of every five citations point at an independent or group practice domain, and the regional-magazine pipeline is effectively closed at 0.7%.
Atlanta pairs the second-highest practice-owned share with a live magazine pipeline. Atlanta Magazine carries 4.1% of the market on its own, and no hospital system reaches 6%.
Minneapolis–St. Paul splits between the specialty-association bucket and practice-owned domains. Hospital systems fall to 16.3%, eleven points under the corpus baseline.
Phoenix has the corpus’s lowest hospital-system share at 13.7%, and national directories rise to 13.1% to fill the space.
Detroit divides almost evenly three ways – no bucket clears 30% – with Henry Ford alone carrying 14.6% of the market.
St. Louis is hospital-dominant at 43.1%, with BJC alone at 12.4%. It produced no regional-magazine citations at all.
Dallas–Fort Worth is one of the corpus’s flattest markets. Practice-owned leads at 27.6%, but no single practice domain reaches even 1.1% – the share is spread thin rather than concentrated.
Denver is the closest hospital-versus-practice call in the corpus: the two sit half a point apart, at 27.5% and 27.0%.
Baltimore runs through Johns Hopkins, which carries 13.9% of the market by itself. Lab and imaging locators reach 7.2% here, the corpus’s highest, and no regional magazine was cited.
Washington splits its top three almost evenly – hospital systems, practice-owned domains and specialty associations within six points of each other – and Washingtonian’s Top Doctors list carries 5.1% on its own.
Houston runs through Houston Methodist at 13.9%, but Kelsey-Seybold is the corpus’s largest single practice-owned domain at 5.7% – a group practice holding hospital-scale ground.
Charlotte is the only market where review aggregators reach double digits, at 12.7%. SouthPark Magazine carries another 5.3%.
Chicago is hospital-led at 33.6%, six points above the corpus baseline, and its specialty-association share falls five points below it.
San Diego pairs a dominant health system with a live magazine pipeline. Scripps carries 16.9% of the market, and San Diego Magazine another 7.3%.
San Francisco has the corpus’s heaviest single-domain concentration: UCSF alone carries 19.3% of the market, and hospital systems together take 40.4%.
Seattle has the corpus’s largest magazine pipeline at 9.2%. Seattle Magazine alone carries 6.3% – six times the market’s biggest practice-owned domain.
Los Angeles is hospital-led at 34.7% with practice-owned down at 16.3%, and its regional-magazine pipeline is all but absent at 0.1%.
Philadelphia sends 38.7% into the specialty-association bucket. Philadelphia Magazine’s Top Doctors list carries 5.0% – more than a third of everything practice-owned domains earn here.
Boston sends nearly half its citations into the specialty-association bucket and only 11.1% to practice-owned domains, second-lowest in the corpus. Boston Magazine is a larger single source here than any hospital system.
New York is the hardest market in the corpus for an independent practice to be cited in. Practice-owned falls to 10.5%, the specialty-association bucket and national directories take 47.8% between them, and no regional magazine was cited.
Exploratory secondary analysis of an existing corpus, single-platform (ChatGPT), cross-sectional, US-only. The 33,340 citation events span 15,283 distinct URLs across 10 specialties and the 20 metros sampled; the pre-registered hypothesis tests were trimmed to 16 retained metros for budget, but the citation inventory keeps all 20. This is the citation-side view of the study; the mention-side analysis, including the “named zero of 200” finding, is reported separately. Source-type buckets were generated by a deterministic Python classifier on May 21, 2026 – the previous "practice-or-other" residual was split into clean buckets using deterministic rules plus Claude Code subagent classification of the long tail. No causal claims; all numbers are descriptive citation-share data. Dentistry is a post-hoc, single-vertical addition collected on June 14, 2026 with model gpt-5.5-2026-04-23 (517 responses across 16 metros and 14 patient-intent templates). It is selectable above for comparison and its own corpus has a dedicated page, but its citations are not counted in the 33,340 total. Full methodology and pre-registered hypothesis verdicts available on request.
The analysis lives in MGMA – including the magazine-pipeline geography (which eight metros carry it, which five have no pipeline) and the three on-Monday moves administrators can run themselves. Read the full article in MGMA →
Also covered in Medical Economics: “Why ChatGPT favors hospital systems” →Run a free scan to see your A–F visibility grade and how often ChatGPT cites your practice.