Before you pay for AI visibility for your practice, audit these five sources

The first article in this series showed why hospital-affiliated physicians dominate so many ChatGPT recommendations: hospital and academic rosters were the largest citation category in our corpus. The second article showed that when ChatGPT does name an independent practice, a third party usually supplies the description. Both trace to the same constraint: a practice can have outstanding clinical credentials, and a search system will still know only what the internet says about it.

Five sources carried most of the citations behind those answers: the practice website, the specialty-board or association profile, the state license record, the regional Top Doctors page and the hospital or academic roster. A citation is a source ChatGPT pulls from to build an answer; a mention is the practice name the patient actually sees. Together those five accounted for roughly four in five of the 33,000 citations we classified.

That is not where most practices spend first. Review sites accounted for 3.8% of the citations in our study, and no citation URL in our corpus pointed directly to a Google Business Profile. Both still matter for local search. But those were not the sources ChatGPT cited most often when describing physicians.

These are not simply the five largest categories; national directories, at 8.3%, outrank two of them. They are where citation frequency and a practice's ability to act overlap. The order below runs by control and applicability, not citation share.

1. Publish the details patients actually ask about

Start with the practice's own website, the only one of the five a practice fully controls. A generic services page tells ChatGPT that a practice offers knee surgery. A dedicated procedure page tells it which conditions the surgeon treats, how the procedure works, who is a candidate, and what training the surgeon brings. When ChatGPT cited a practice website in our study, the description it produced was only as specific as what that practice had published.

Once those pages carry real information, schema markup can label the physician, procedure and credentials explicitly. Treat it as labeling, not a substitute for substance.

The website is also the reference copy for the rest of this audit: use the same practice name, address, phone number, credentials, and procedure language when you claim or correct the outside profiles below.

2. Correct your specialty-board and association profiles

Specialty associations and credentialing pages are the largest third-party category most practices can review directly, at 22.1% of the citations in our study. For each physician, check the name, current practice name, address, phone number, website, board certification, fellowship training and current affiliations.

A stale address left by a former employer was the most common profile error we encountered. Consider a cardiologist who leaves a hospital, opens an independent practice across town, and updates his website. If his specialty-board profile still shows the hospital's address, ChatGPT can find it there and give it to a patient years later. Updating the practice website does not resolve that, because the page ChatGPT cites is somewhere else.

How you fix these profiles can vary. Certification Matters takes its data from the ABMS member boards rather than from physicians directly, and specialty societies maintain their own directories. Expect changes to take time to appear, and ask the former employer to update or remove the old bio page.

3. Check your state medical-board record

State medical boards provide public physician-license lookups, and those pages accounted for 5.1% of the citations in our study. What each board displays varies, so verify whatever yours publishes: name, license status, any public address, specialty or certification fields, and the disciplinary and administrative entries. How you fix these differ by state, and the board's own site will name the office that handles them.

4. Find your region's Top Doctors process

The national ranking brands are the better-known names, but regional magazine pages produced substantially more citations in our corpus, including lists published several years earlier. Those lists run through processes that vary by market, sometimes using Castle Connolly and sometimes a magazine's own nomination cycle.

They are also metro-specific. In our sample, San Diego, Seattle, Washington and Boston produced repeated citations, while New York and Los Angeles produced almost none. Identify the publication serving your metro, find out who administers its list and whether nominations are peer-based, confirm your physicians are eligible,and put the nomination deadline on the annual marketing calendar.

5. Update your hospital and academic rosters

Hospital and academic roster pages were the largest source category in our study, at 27.5% of citations, but they come last here because they apply only to physicians who already hold privileges or an academic appointment. Confirm the name, specialty, practice or group name, office location, phone number, website, and privilege or faculty status. A practice cannot edit these pages, but the physician or an administrator can flag errors to the hospital's medical-staff office or web team.

What this audit can and cannot change

Being cited is not the same as being mentioned, and being mentioned is not the same as being chosen by a patient. Citation frequency also does not show where a practice ranks in local search. Reviews and a well-maintained Google Business Profile influence local visibility and help patients evaluate a practice after finding it. Our study covers ChatGPT only, in May 2026, and is correlational. Correcting a profile will not overcome a structural disadvantage in specialty, market or affiliation. The data supports an order of operations, not a guaranteed lift.

Start by auditing all five. Correct factual errors immediately, then expand thin website content and calendar the recurring checks and nomination cycles. Auditing the outside profiles for one physician may take an afternoon; getting an outside organization to act on a correction takes longer and publishing substantive website content is ongoing work.

When a patient uses ChatGPT to narrow their options, a practice left out of the answer may never be considered. Practices cannot choose which sources ChatGPT cites, but they can make sure the sources within reach reflect the practice as it exists today. Keeping profiles current and publishing specific website content gives physicians a better chance to be found and understood on the strength of their own credentials, rather than an old employer’s page or an incomplete listing.

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

Companion: See which sources ChatGPT cites, and how the mix shifts by specialty, on our citation map.

Methodology

The figures come from roughly 33,000 ChatGPT citation events, generated by about 3,400 responses to roughly 1,200 patient-style prompts across 10 specialties and 20 metros, with each prompt run several times to absorb model variance. Every cited URL was classified by source type; the shares here are of all citation events in that corpus. This is a separate dataset from the random sample described in the first article, in which 200 practices drawn at random from the NPPES registry were tested and none were named. Because the source map covers the practices and institutions ChatGPT actually surfaced, it carries a selection bias toward them. All queries ran against GPT-5.5 through the OpenAI API in May 2026. The study covers one platform and the United States only; Perplexity, Gemini, and Google's AI Overviews may behave differently. The findings are correlational, and metro-level results are illustrative rather than exhaustive.