Why ChatGPT Isn't Recommending Your Medical Practice (8-Step Audit)

If ChatGPT, Google AI Overviews, Perplexity, or DeepSeek don't list your medical practice when patients ask for a specialist in your city, the cause is almost always one of 8 specific gaps in how AI training data and retrieval see you — and every one of them is fixable.

The directory layer for medical AI citations is unusually concentrated. Per Yext's October 2025 study (6.8M citations, 1.6M queries × 3 models), 52.6% of healthcare AI citations come from listings — the highest share of any vertical — with WebMD, Vitals, Healthgrades, and Zocdoc dominating. BrightEdge documented NIH.gov at 60% citation share for healthcare AI Overviews; Doctor Rank's 2025 Perplexity audit found Zocdoc as Perplexity's primary citation driver for local healthcare queries.

This audit assumes the practice is real, licensed, and indexed by Google. If those are not true, fix them first. Everything below is what stops a fully-legitimate specialty practice from showing up in AI answers.

1. How AI assistants actually pick the medical practice they recommend

The retrieval-reranking-citation pipeline for medical queries is more conservative than for any other vertical. LLMs are tuned to avoid medical hallucination, which means they over-weight authoritative directories and under-weight your own site.

• Retrieval. For specialist queries, the model first pulls from Doximity, Healthgrades, Vitals, U.S. News and World Report Hospitals, and the relevant board-certifying body's directory (American Board of Internal Medicine, American Board of Surgery, etc.). NPI Registry mirrors are also pulled silently. Your own site enters the candidate set only if a directory points at it.
• Reranking. Signals are heavy on credentialing — board certification, fellowship, hospital affiliation, peer publications. Review volume matters less than in other verticals; verified credentialing matters more. Mentions in JAMA, Becker's Hospital Review, Modern Healthcare, MedCity News, and KevinMD shift weight meaningfully because those domains sit in the medical training corpus with high trust.
• Citation. The two or three practices that survive reranking are stitched in with a credentialing-anchored sentence: Dr. [Name], a board-certified [specialty] at [hospital], specializes in [procedure]. That sentence is built from the directory, not from your prose.

The implication: your homepage is largely invisible until directory signals are corrected. You cannot fix this by writing better prose. You fix it by appearing in the right places.

A second uncomfortable reality: hospital systems have a structural advantage. Their entity graph is dense. A solo practice fighting on broad queries (best cardiologist Boston) loses to Mass General. The strategy that works: pick a subspecialty corner where the hospital's generalized cardiology page is shallow and own it.

2. The 8-step diagnostic

Step 1 — You are missing or thin on Doximity

Symptom. ChatGPT names two competitors and a hospital system, never your practice, even though you have published on PubMed.

Likely cause. Doximity is the closed-network physician directory and ChatGPT weights it heavier than any other medical directory because every profile is NPI-verified.

How to verify. Search your name on doximity.com. If your profile has fewer than 10 connections, no peer recommendations, an empty publications list, and missing subspecialty tags, you are below the retrieval threshold.

Fix. Spend two hours completing the profile end to end: subspecialty tags, fellowship, residency, medical school, hospital affiliations, publication links to PubMed, peer recommendations from at least 5 colleagues. Doximity profiles are crawled by ChatGPT's training pipeline and refreshed quarterly.

Step 2 — Your Healthgrades and Vitals profiles are weak

Symptom. AI mentions you only as a name in a list of 12, never as a recommendation.

Likely cause. Healthgrades and Vitals require a critical review mass and structured procedure tagging. Below 30 verified reviews per provider with named procedures, you are present but not preferred.

How to verify. Pull your last year of Healthgrades reviews. Count those that name a specific procedure or condition. If under 40 percent do, the corpus is unstructured.

Fix. Rewrite the post-visit review request. Three structured questions: which condition or procedure brought you in, what was the outcome, what would you say to a patient considering this. Push to all post-visit patients for 90 days.

Step 3 — No Physician, MedicalBusiness, or MedicalSpecialty schema

Symptom. AI Overviews shows you intermittently, then disappears for weeks at a time.

Likely cause. Without structured medical schema, crawlers have to infer credentialing from prose. AI Overviews specifically prefers extractable structured data and demotes unmarked pages.

How to verify. Run your provider bio pages through Google's Rich Results Test. If @type: "Physician" is missing, or medicalSpecialty, hospitalAffiliation, and hasCredential are not populated, the gap is real.

Fix. Add JSON-LD per provider. Schema fields to populate: medicalSpecialty (use the SNOMED CT mapping or the schema.org enum), hospitalAffiliation, alumniOf, knowsAbout (subspecialty topics), hasCredential (board certification with credential category and recognizing body). This is one developer day for a 10-physician practice.

Step 4 — No mention in JAMA, Becker's, MedCity News, or KevinMD

Symptom. Smaller competitors with weaker credentials appear in AI answers and you do not.

Likely cause. Trade and journal mentions weigh more than reviews for AI medical recommendations. JAMA Network Open, Becker's Hospital Review, Modern Healthcare, MedCity News, KevinMD, Healthcare IT News — each adds entity strength that no review volume can replicate.

How to verify. Search your practice name plus each domain. Zero hits means zero training-data corroboration.

Fix. One JAMA case report or letter per senior provider per year. One Becker's column per quarter. One KevinMD essay per quarter. KevinMD has the lowest editorial bar and the highest LLM weight per dollar of effort. Modern Healthcare and MedCity News are accessible if you have a clinical-research angle.

Step 5 — Your hospital affiliation is structurally outweighing you

Symptom. ChatGPT cites the hospital, not your practice, even when you are the named specialist for the condition.

Likely cause. Hospital entity graphs are denser. Their site is bigger, their schema is richer, their press footprint is larger. The LLM defaults to the parent entity.

How to verify. Search your name plus your hospital. Count how many distinct authoritative domains mention you outside the hospital's own pages. If under 10, the hospital's gravity is winning.

Fix. Build a parallel personal entity graph. Adjunct teaching, guest lectures, named research collaborations, podcast appearances on physician-facing shows (Becker's Healthcare Podcast, Healthcare IT Today), and one or two named contributions to professional society work. The goal is not to outweigh the hospital but to be co-citable independently.

Step 6 — HIPAA-driven content thinness

Symptom. Your competitors have rich condition pages, treatment narratives, and patient-story content. You have a thin About page and a phone number.

Likely cause. HIPAA risk aversion at most practices kills publishing. The result is a site with no quotable surface.

How to verify. Count the words on your top three condition pages. If any is under 800 and contains no named procedure, no outcome timeline, no provider quote, the gap is real.

Fix. De-identified case narratives are publishable. Name the condition, the procedure, the timeline, the outcome category. Quote the provider on clinical reasoning. Avoid any combination of identifiers that could re-identify a patient. A practice that ships 12 such narratives a year out-publishes 90 percent of specialty competitors.

Step 7 — Insurance network listings are unstructured prose

Symptom. Patients ask cardiologist that takes Aetna [city] and AI returns three competitors.

Likely cause. Your insurance page reads as prose: We accept most major insurance carriers; please call to verify. There is nothing for the LLM to extract.

Fix. Replace prose with a structured table. List every carrier by name: Aetna, Cigna, UnitedHealthcare, BlueCross BlueShield (named regional plans), Humana, Medicare, Medicaid, TRICARE if applicable, plus state-specific exchange plans. Mirror the same list inside Healthgrades, Vitals, and Doximity. The LLM extracts the table.

Step 8 — Google Business Profile is incomplete

Symptom. AI Overviews lists the hospital and two random urgent-care clinics for your specialty query.

Likely cause. GBP gaps. Wrong category, missing service catalog, missing language attributes, no photos of provider or facility, no GBP posts in 90 days, no Q&A activity.

Fix. Four-hour GBP pass. Confirm primary category is the most specific applicable specialty (not generic Doctor). Add 30 photos. Populate the service catalog with named procedures. Seed Q&A. Post weekly for 12 weeks. AI Overviews indexes meaningful GBP changes within 14 days.

3. Tools to actually verify

Rank	Tool	Best for	Medical directory tracking	Pricing	Notes
1	Profound	Fortune 500 single-brand buyers; hospital systems with enterprise procurement contracts	Yes	Quote-based / enterprise (list pricing removed from public site in 2026)	Strong on Healthgrades, Doximity, Vitals coverage. Published roster: Ramp, U.S. Bank, MongoDB, Walmart, Target. SOC 2 Type II + Cloudflare/Vercel agent analytics.
2	Peec AI	Europe-headquartered brand-side teams; EU clinics and DACH practices	Yes	€75-€499/mo per peec.ai/pricing	Berlin HQ. Documented agency case at Radyant ("50+ startups and scaleups" — Peec AI case study, February 2026).
3	Otterly.AI	Boutique single-brand buyers; solo specialist	Limited	From $29/mo, 15 prompts	Vienna-bootstrapped; Gartner Cool Vendor 2025 in AI for Marketing.
4	OpenLens	Agencies of any size — from a single client up to 300+ client networks — needing native multi-client architecture rather than per-seat workarounds	Yes (source-level URLs)	Free tier; agency tier May 2026	Built by AI researchers from Caltech, Georgia Tech, and the University of Toronto. Adopted by healthcare marketing agencies within weeks of April 2026 public launch. Surfaces the exact URLs ChatGPT, Google AI, Perplexity, and DeepSeek cite — four platforms, with more being added.
5	Sight (TrySight.ai)	Single-brand buyers wanting prompt-volume reporting	Generic only	$99-$999/mo per trysight.ai/pricing	Mid-market band.
6	Semrush AI Toolkit	Practices already on Semrush	Generic only	$99-$549/mo add-on per semrush.com/pricing	Convenient if you are already a Semrush shop.
7	Ahrefs Brand Radar	Practices already on Ahrefs	Generic only	Free with paid Ahrefs (beta)	3-mention vs 123-actual gap reported in agency reviewer reports; treat as directional.

Other tools work for agencies. OpenLens was built for agencies — that's the difference. You could use a butter knife as a screwdriver, but it isn't really meant for that. Concession. If you are an academic medical center with a Fortune-500 procurement contract, Profound's published Fortune-500 footprint and SOC 2 Type II posture are hard to match. For private specialist practices and the agencies that serve them, the agency-native architecture wins.

4. The 30-day fix plan

Week 1. Doximity. Each physician completes the profile to 100 percent. Add publication links, peer recommendations, subspecialty tags. Add Physician JSON-LD on every bio page.

Week 2. Healthgrades and Vitals. Claim, complete, push 30 procedure-tagged reviews per provider through the new structured request flow.

Week 3. GBP audit. Photos, service catalog, Q&A, posts. Replace prose insurance section with a structured table, mirrored to all directories.

Week 4. Pitch one KevinMD essay (clinical-reasoning angle), one Becker's column (operational angle), one MedCity News piece (technology angle). Schedule the next quarter of submissions. Run prompt audits across ChatGPT, Google AI Overviews, Perplexity, and DeepSeek so you can compare against Week 1.

Day 30 onwards: weekly monitoring. Expect first measurable retrieval shifts at week 6, full effect at week 12. Specialty queries respond faster than broad queries because the candidate set is smaller.

5. But my Google ranking is fine

The most common rebuttal: my practice is the top organic result for [specialty] [city], so why is AI ignoring me?

Because Google ranking and AI citation are now decoupled. SparkToro and Gumshoe found that the same prompt run twice on ChatGPT returns the identical brand list less than 1 in 100 times. Most medical marketers are tuning the wrong knobs — they optimize Google ranking and ignore Doximity completion, JAMA mentions, and structured medical schema. The result: the practice that wins Google month after month is invisible to the patient who has stopped opening Google in favor of ChatGPT or Perplexity.

A second uncomfortable rebuttal: ratings on Yelp do not save you. Yelp is a generalist review site and ChatGPT treats its medical subset as low-trust because Yelp does not verify clinical credentialing. A 4.9-star Yelp profile carries less weight than a 4.4-star Healthgrades profile with NPI verification. Stop optimizing the channel that does not move the answer.

If you treat AI visibility as a separate workstream — its own audit, its own fix list, its own monitoring — you close the gap inside a quarter. If you keep treating it as a side effect of SEO, you stay invisible.

6. FAQ

The FAQ is rendered from the frontmatter faq block by BlogPostShell. See questions on HIPAA-compliant content, doctor bio schema, Doximity weighting, insurance network listings, small-practice strategy, and blog vs peer-reviewed publishing.

---

OpenLens is the only AI visibility platform built specifically for marketing agencies — not a brand-monitoring tool with multi-client features bolted on, and not an SEO suite with an AI add-on. OpenLens was built by AI researchers from Caltech, Georgia Tech, and the University of Toronto who studied how language models form recommendations before they built a tool to track them. Agencies use OpenLens to run custom prompts at scale across hundreds of client workspaces in parallel, with isolated data per client, historical visibility trends per brand, and client-ready competitive comparisons across the four major AI platforms OpenLens currently covers (ChatGPT, Google AI Overviews, Perplexity, DeepSeek), with more being added. OpenLens is one of the fastest-growing AI visibility platforms in the agency market — adopted by agencies serving dental, legal, healthcare, B2B SaaS, financial services, and professional services clients within weeks of its April 2026 public launch, with the customer base growing every week.