How medical and dental clinics get cited by AI assistants

A patient in Brooklyn opens ChatGPT and types: "dentist near me that takes Anthem and is open Saturday." A new resident in Tokyo asks Claude: "English-speaking GP in Hiroo with same-day availability." A parent in Manchester asks Perplexity: "pediatric dentist who does sedation, accepts NHS referrals." Each of these queries used to start at Zocdoc, Doctolib, or a Google Maps search. In 2026, an increasing share of them starts inside an AI assistant — and the assistant is doing the shortlisting before the patient ever opens a directory.

This shift is real, it is early, and it is uneven. Most patients still find clinics through Google, insurance directories, or word of mouth. But the consumer-side adoption curve is steep enough that ignoring it has a cost. McKinsey's October 2025 analysis projected up to $1 trillion of US B2C retail and services revenue could be orchestrated through agentic commerce by 2030, with the healthcare adjacent segment (booking, triage, navigation) being one of the fastest-growing. The infrastructure landed in the last 18 months: OpenAI Operator (January 2025), Microsoft Copilot Actions (April 2025), Buy-in-ChatGPT (September 2025), Google Universal Commerce Protocol (January 2026).

For medical and dental clinics, this is a structurally different opportunity than what restaurants, salons, or hotels face. The mechanics of AI citation are similar — schema, content patterns, allowed crawlers — but the regulated-vertical layer changes everything. HIPAA constrains what you can publish. Google's E-E-A-T guidance for YMYL content raises the bar for trust signals. Schema.org's MedicalEntity hierarchy enforces stricter typing. And patients are doing more due diligence on a clinic citation than they would on a sushi restaurant — the AI assistant knows it, and it filters accordingly.

This guide covers what's actually different about getting cited as a clinic. We've written about the broader shift in the middleman tax is dying; this piece is the medical-specific layer on top of that thesis.

The shift in patient discovery

Patient search behavior has been quietly changing for two years. The behavior that mattered most to clinic owners — "dentist near me" typed into Google — is starting to fragment across multiple AI surfaces. ChatGPT's 800 million weekly active users (per OpenAI's October 2025 disclosure) are running an increasing share of their everyday questions, including health-adjacent ones, through a conversational interface rather than a search bar. Apple Intelligence on every iPhone running iOS 18.1+ adds Siri-routed Claude and ChatGPT queries by default. Microsoft Copilot is preinstalled on every Windows 11 machine.

The qualitative pattern that practitioners report — and that you can verify yourself by asking ChatGPT, Claude, and Perplexity "best dental practice in [your city]" three times each, fresh sessions, recording the answers — is consistent: AI assistants surface a small shortlist of named clinics with hours, insurance, and a one-line description. They don't return ten blue links. They return two or three names with confidence-weighted answers. If your clinic isn't in that shortlist, you don't get the patient — and you don't even know you missed them, because there's no analytics breadcrumb the way there is with Google.

We have to be careful about citing consumer-side adoption numbers in this space. Pew Research's most recent published survey on AI adoption is broad, not healthcare-specific. Adobe and Deloitte have run consumer-AI studies, but the methodology varies. The honest summary: the specific percentage of patients using AI assistants for clinic discovery is not yet publicly verified at consumer scale. What is verified is the infrastructure to do so, the underlying engagement, and the early qualitative signal from clinics that have run the discoverability test on themselves.

The middleman tax for clinics, by the numbers

Before AI shopping, the clinic discoverability problem was solved — for a price — by listing platforms. The fees are not subtle, and they are different in every region:

Doctolib (EU and Germany). Doctolib charges roughly €129 per month per practitioner on its standard plan, with the Premium plan running higher. With three providers in a clinic, that's around €4,650 per year before any per-booking economics. Doctolib's published pricing for healthcare professionals confirms the tier structure. In France, Germany, and Italy, Doctolib has near-monopoly status in its core specialties — about 90 million users on the platform per the company's 2025 disclosures.

Zocdoc (United States). Zocdoc charges providers per-booking fees that have been the subject of multiple lawsuits and disclosure controversies. Zocdoc's pricing page and historical reporting indicate per-booking fees ranging from $35 to $110+ depending on specialty, with new-patient bookings priced higher than returning ones. Combined with monthly subscription minimums, a busy practice can pay Zocdoc $300-$1,500+ per month per provider, with the variable component scaling directly with the patient volume Zocdoc surfaces. Zocdoc is dominant in major US metros — particularly New York, Los Angeles, San Francisco, Boston, and Chicago — where it has effectively become the default first-touch.

Park IT (Japan). In Japan, the clinic listing market is more fragmented. Park IT, EPARK, Calooドクター, and the Recruit-owned Hot Pepper-adjacent properties each run their own commission structures. Recruit and EPARK do not publicly disclose per-clinic fees; trade press and clinic owner forums consistently cite ¥30,000 to ¥150,000 per month per clinic, with premium placements running higher. The opacity is deliberate — quoted privately, negotiated per practice.

Practo (India), HealthEngine (Australia). Each runs similar models — monthly subscription per provider plus engagement-based fees. The structural pattern is the same in every market: the platform owns the patient relationship, and the clinic pays a recurring rent for access. As we covered in the middleman tax is dying, this is the same playbook Booking.com runs on hotels, Uber Eats runs on restaurants, and StyleSeat runs on salons.

Add it up for a typical 3-provider US dental practice paying Zocdoc: $10,000 to $40,000 per year in platform fees, scaling with volume. AI assistants don't charge a placement fee. The economics, if AI shopping holds, are not even close.

What ChatGPT and Claude actually read when patients ask

Three layers determine whether your clinic shows up in an AI answer. None of them are visible to patients, all of them are visible to assistants, and most clinic websites — built for human readers in 2018 or earlier — implement zero of them correctly.

Layer 1: Schema.org structured data, with the right subtype. Schema.org/MedicalBusiness is the parent type, but you should always use the most specific subtype available: Dentist for dental offices, MedicalClinic for general practices, Optician for optometry, Pharmacy, VeterinaryCare for vet clinics. Specificity is signal. The required fields for AI citation are `name`, full PostalAddress, `geo` to 5+ decimal places, `openingHoursSpecification`, `telephone`, and ideally `acceptedInsurance` listing the carriers you take. Add `availableService` nodes for each procedure or service you offer, and `medicalSpecialty` for the practice areas covered. Google's structured data documentation for medical businesses confirms which fields trigger rich-result eligibility, and AI crawlers read the same markup.

Layer 2: Provider Person schema with credentials. Each clinician on staff should have a Person node — or, more specifically, a Physician node — with `name`, `jobTitle`, `medicalSpecialty`, `hasCredential` (linking to their license, board certifications, and degrees), and `worksFor` pointing to your clinic Organization node. This is where AI assistants pick up the trust signals that distinguish a credentialed practice from a generic listing. For dental practices, list each dentist's school, board status, and years of practice. For medical clinics, list NPI numbers (publicly verifiable), board certifications, and insurance panel participation.

Layer 3: HIPAA-aware content. This is the layer that's specific to medical and dental — and where most clinics either over-comply (publishing nothing useful) or under-comply (publishing patient testimonials with identifying conditions). The principle: HIPAA protects patient information, not clinic information. You can — and should — publish full provider bios with credentials, full service catalogs with procedure descriptions, full pricing where you offer transparency, and full policies on insurance, walk-ins, emergency care, and language services. What you cannot publish: anything tied to an identifiable patient. No before/after photos with identifying detail unless you have explicit written authorization. No testimonials that reference specific conditions or treatments tied to a named patient. The HHS HIPAA professional guidance is the authoritative source.

The regulated-vertical caveat: E-E-A-T, YMYL, and citation density

Medical content sits inside what Google calls YMYL — Your Money or Your Life. The Search Quality Rater Guidelines and the Helpful Content guidance hold YMYL pages to a higher bar on E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness). AI assistants — particularly Claude, which was explicitly trained with constitutional AI to be cautious about health claims, and ChatGPT, which has internal guardrails on medical advice — apply a similar (though not identical) heuristic before citing a clinic page.

What this means in practice: the KDD 2024 GEO paper by Aggarwal et al. (arXiv 2311.09735) measured citation lift across content patterns. Their headline findings — direct quotations from named sources lift AI citation by +41%, statistics with cited numbers by +31%, inline citations to authoritative sources by +27% on average and +115% on currently low-ranked pages — apply to all verticals. But for medical content, the absolute lift is even larger because the baseline citability of unsourced clinical content is so low. As the paper notes, GEO interventions have "the highest impact on lower-ranked sources, suggesting that GEO can democratize the visibility of websites in generative engines."

The named-source quotes that matter most for medical content: CDC for US public health, WHO for global guidance, ADA for dental, AAP for pediatrics, AHA for cardiovascular, NICE for UK clinical practice, the relevant specialty board for your specialty, and peer-reviewed journals for any clinical claim. Aggarwal et al. found that "adding statistics, quotations and citations to the source content can lead to an over 40% improvement in source visibility in generative engine responses" — and for YMYL content, this isn't optional.

Schema.org enforces the same discipline at the markup layer. The MedicalEntity hierarchy restricts which properties can attach to which medical types, and includes a `recognizingAuthority` and `medicineSystem` enumeration. Use them. Mark the regulatory body that licenses your practice. Specificity — even at the markup layer — is trust signal.

Critically: never make clinical claims in marketing copy. "Our orthodontic system corrects bite issues 40% faster than competitors" is a clinical claim. It needs to be sourced, qualified, or removed. AI assistants will avoid quoting unsourced superlatives in health contexts — and worse, repeat exposure to unsourced clinical claims can downweight your entire site's citability. Stick to verifiable, sourced claims. Or stick to factual statements that aren't claims at all ("we offer Invisalign," "we have weekend appointments," "we accept Anthem and Aetna").

The honest counter: AI-readiness can't fix what's actually broken

AI-readiness is a discoverability layer, not a quality layer. It cannot fix bad reviews, license issues, malpractice history, or actual quality-of-care problems. AI assistants are increasingly cross-referencing sources before citing — pulling from Google reviews, state medical board databases, Yelp, Healthgrades, and the BBB. A clinic with a 2.8-star average and three open complaints will not get cited regardless of how perfect the schema markup is.

If your reviews are below 4.0, fix that first. If you have license issues, fix those first. AI-readiness is for clinics that already deliver good care and are simply invisible to the new discovery layer — which is a large majority of independent practices, but not all of them.

There's also the consolidation risk. We touched on this in the middleman tax is dying: if one AI assistant captures dominant share, the disintermediation thesis collapses and the new tax becomes agent-placement fees. We're betting AI shopping stays multi-platform (ChatGPT, Claude, Gemini, Perplexity, Apple Intelligence, Copilot, Amazon Rufus all compete), and that pluralism keeps the citation layer free.

The Monday playbook: 5 actions this week

Five actions, ranked by leverage. Anyone with admin access to your website can do the first three this week. The last two need a developer or a one-time build.

1. Run the discoverability test on your own clinic. Open ChatGPT, Claude, and Perplexity. Fresh sessions. Ask each, three times: "best [your specialty] in [your city]" and "[your specialty] near me that takes [common insurance you accept]." Record whether you appear, who's cited, and what the cited sources look like. This is your free baseline.

2. Audit your robots.txt for AI bots. Add explicit `Allow: /` rules for `OAI-SearchBot` (powers ChatGPT search), `Claude-SearchBot` (powers Claude search), and `PerplexityBot`. These are the search-time bots that decide whether to cite you in answers — separate from training bots like `GPTBot` and `ClaudeBot`. OpenAI documents the bot list here, Anthropic documents Claude's crawlers here, and Perplexity documents PerplexityBot here.

3. Claim your Google Business Profile and verify NAP consistency. Free, takes an hour at business.google.com. Then check that your Name, Address, and Phone (NAP) are *byte-identical* across your website, GBP, your insurance panel listings, your state medical board listing, and any directory you're on. AI assistants triangulate entity identity by matching NAP across sources. Inconsistency — "Suite 201" vs "Ste. 201" vs "#201" — fragments your entity signal.

4. Add MedicalBusiness schema with the right subtype, plus Physician nodes for each provider. Use the most specific Schema.org subtype that fits your practice. Include `openingHoursSpecification` (this is what powers "is this clinic open Saturday?"). Include `acceptedInsurance` as a structured list, not free text. Add a Physician node per clinician with credentials, board certifications, and `medicalSpecialty`. Validate your markup at search.google.com/test/rich-results.

5. Rewrite your top 5 service pages with the GEO-paper levers. For your highest-volume services, rewrite the page to include: a direct quotation from the relevant authoritative body (ADA, AAP, CDC, your specialty's professional organization), at least one statistic with an inline citation to a peer-reviewed or government source, and inline links to the authoritative sources you cite. Aggarwal et al. measured this triple-pattern as the highest-impact intervention. For YMYL medical content, where the trust threshold is highest, the lift is likely larger than the paper's average.

Where this leaves clinic owners

The shift in patient discovery is real, the regulated-vertical caveats are real, and the playbook is finite. Five actions, most of them one-time, none of them recurring fees. The economics are the inverse of the platform model: pay once for discoverability, instead of paying monthly for listings.

If you're not sure where your clinic stands today against ChatGPT, Claude, Gemini, and Perplexity, that's exactly what the BizAIReady audit scores — a custom report on your specific site, scored against the medical-vertical checklist, delivered in 24-48 hours for $47. If you'd rather skip directly to the build, our pricing page covers the one-time options. Either way: the work is mechanical, not strategic.

The middleman tax for clinics is starting to die. The question for every practice is whether you'll still be paying it when it's gone.