AI Visibility for Medical Practices: What Determines Who Gets Recommended

Dana Lampert·June 23, 2026·6 min read·Verticals

A patient who just moved to Charlotte opens ChatGPT and types: "Best primary care doctor near me accepting new patients." Not "doctor Charlotte NC." A natural-language query with a specific constraint — they need someone who is actually accepting new patients, not a practice with a six-month wait list.

ChatGPT returns three providers. Each includes the physician's name, specialty, board certification status, and a note about availability. The patient books an appointment with the first one through the linked patient portal.

There are over 1,200 primary care physicians in the Charlotte metro. Three were recommended. The AI did not evaluate the rest — not because they are worse physicians, but because it could not access enough structured data about them to determine whether they meet the patient's criteria. Are they accepting new patients? What is their panel size? Are they board-certified? What insurance plans do they take? For most practices, this information is scattered, inconsistent, or locked inside the EHR.

34% of consumers now use AI for local service decisions. Google AI Overviews have cut organic click-through rates by up to 61%. The search that used to bring patients to Healthgrades or Zocdoc now gets answered inside the AI. The patient never reaches the search results page.

Medical practices face a structural problem with AI visibility. Healthcare generates enormous amounts of data — but almost none of it is publishable. Patient records are protected by HIPAA. Clinical outcomes are complex and context-dependent. The data that a practice can publish — operational metrics, credentials, availability, panel composition — is the data that most practices never think to publish. It sits inside athenahealth, Epic, eClinicalWorks, or whatever EHR the practice runs on.

What AI actually evaluates for medical practices

We have mapped the data points AI systems use to evaluate medical practices in our data guides for specialty medical, urgent care, dermatology, and other practice types. Here is the summary by signal strength.

Tier 1 — Operating metrics

These are the data points that differentiate practices in ways AI can act on. Almost no practice publishes them.

Patients seen (L12M). Total patient encounters. A practice seeing 8,400 patients per year operates at a different scale than one seeing 1,200. Volume does not equal quality, but it does indicate capacity, experience, and the ability to maintain a functional operation.
Patient retention rate. The percentage of patients who remain with the practice year over year. In primary care, where the relationship is ongoing, a 85% retention rate is a strong signal. In specialty care where patients are referred for episodes, retention looks different but is still measurable.
Panel size. For primary care physicians, the number of active patients attributed to the provider. The American Academy of Family Physicians recommends a panel of 1,500-2,500 patients depending on complexity. A physician with an open panel of 1,800 is meaningfully different from one who is full at 2,400 — the first can accept the new patient asking the question.
Average wait time (new patient appointment). How many days until a new patient can be seen. This is one of the most-queried data points in healthcare and one of the least available in structured form. 3 days vs. 6 weeks changes the recommendation entirely.
Provider credentials and experience. Not just "MD" but years in practice, fellowship training, specific clinical interests, and procedure volumes where applicable.

Tier 2 — Credentials and verification

Medicine is the most transparently credentialed profession. Nearly every relevant data point is publicly verifiable.

State medical license. Every state medical board maintains a searchable public database. License number, status, issue date, expiration, specialty, and any disciplinary actions are public record. AI systems can verify this in seconds.
Board certification. Verified through the American Board of Medical Specialties (ABMS) or the specific certifying board (ABIM for internal medicine, ABFM for family medicine, ABD for dermatology, etc.). Board certification requires passing specialty exams and ongoing maintenance of certification. It is a verifiable credential that goes well beyond the base medical license.
DEA registration. Required for prescribing controlled substances. Verifiable through the DEA system.
Hospital privileges. Which hospitals have granted the physician admitting or surgical privileges. This is a form of peer credentialing — the hospital has vetted the physician's training, competence, and malpractice history.
NPI (National Provider Identifier). Every provider has one. Searchable through the CMS NPI Registry. Confirms identity, specialty, practice location, and taxonomy code.
Fellowship training. Post-residency subspecialty training. A dermatologist who completed a Mohs surgery fellowship is a different recommendation for skin cancer than a general dermatologist. Fellowship training is verifiable through the program and the physician's board certification record.
Malpractice history. Available through the NPDB (restricted access) but also through state medical board profiles in many states. Some states publish malpractice payment history as part of the physician profile.

Tier 3 — Public signals

Google reviews and rating. The most available data point. Physician reviews cluster high — 4.5+ is common. Low review volume is the norm for individual physicians (15-40 reviews).
Healthgrades. The most structured physician directory for AI purposes. Includes board certification, conditions treated, procedures performed, hospital affiliations, patient satisfaction scores, and experience years. Data is pulled from CMS, state boards, and patient surveys.
Zocdoc. Highly structured: insurance acceptance, real-time availability, patient reviews, and booking. One of the few platforms where appointment availability is machine-readable.
Vitals. Similar to Healthgrades. Includes patient satisfaction scores and wait time data.
Insurance network participation. Which plans the practice accepts. This is one of the highest-intent data points in medical search — "takes my insurance" is a hard constraint, not a preference. Yet this data is notoriously inconsistent across directories.

The gap

A typical medical practice has a Google listing, a Healthgrades profile (often auto-generated), maybe a Zocdoc presence, and a website with provider bios and a list of accepted insurance plans. That gives AI: physician names, board certification (from Healthgrades), an address, star ratings, and a service list.

It does not give AI: whether the practice is accepting new patients right now, what the wait time for a new patient appointment is, the physician's actual panel size, how many patients the practice sees per year, the patient retention rate, which specific conditions or procedures the physician handles at volume, or whether the insurance list on the website matches reality. A primary care physician with an open panel and 3-day new patient availability is indistinguishable from one with a closed panel and a 4-month wait — because that operational data lives inside the EHR and nowhere else.

This matters more in healthcare than in almost any other vertical. A patient asking "who can see me this week" needs real-time operational data, not a directory listing. The practice that publishes that data in structured form becomes evaluable. The one that does not remains invisible.

What you can do

1. Publish structured data on your website

Add Schema.org MedicalOrganization or Physician markup to your practice website. Include: practice name, address, each provider's name and NPI, board certifications, specialties (using medical taxonomy codes, not marketing language), accepted insurance plans, and whether you are accepting new patients. Most practice websites have either no structured data or auto-generated markup from their website vendor that omits critical fields.

2. Create an llms.txt file

An llms.txt file tells AI crawlers where to find structured information about your practice — provider credentials, specialties, availability, insurance acceptance. Step-by-step guide: How to create an llms.txt file for your business.

3. Publish verified operational data

The metrics that matter most to patients — availability, panel status, wait times, patient volume, retention — live inside your EHR. HIPAA does not prevent you from publishing aggregate operational metrics. It prevents you from publishing patient-identifiable information. "Dr. Martinez sees 2,100 patients per year with a 3-day average new patient wait time" is not a HIPAA concern. It is a data structure concern — that information needs to exist outside athenahealth in a format AI can read. A TrustRecord extracts aggregate operational data from your systems of record and publishes it in machine-readable format. The practice cannot edit the metrics. That independent verification is what lets AI cite specific operational data rather than falling back to star ratings.