A hospital website should publish four kinds of schema.org markup: Hospital on each facility page, IndividualPhysician on each consultant page, MedicalSpecialty values on both, and FAQPage on pages that answer real patient questions in visible text. Every field in that markup must match the visible page and the Google Business Profile word for word. Google says on the record that AI Overviews need no special schema, so the markup earns its place by making your facts unambiguous and consistent, and it does nothing for a page whose facts are wrong.
Hospital marketing teams often ask the schema question the wrong way round. They ask which tag gets them into ChatGPT. The better question is which facts an engine must read without guessing.
What the engines have said on the record
Google is explicit. Its guidance on AI features says there are no additional requirements to appear in AI Overviews or AI Mode, and that no special schema.org structured data needs to be added (Google Search Central, "AI features and your website", last updated 10 December 2025).
Microsoft has gone further in the other direction. Fabrice Canel of Microsoft Bing told the SMX Munich audience that schema markup helps Microsoft's LLMs understand content (Search Engine Roundtable, 20 March 2025). That is a conference statement, and it names no weighting.
For ChatGPT, Perplexity and Google's Gemini app, we found no published statement on whether their answers read JSON-LD directly, as of September 2026. When these engines search the web, they work from pages a search index has already processed, so markup may reach them indirectly through that index.
So the honest position for a CMO is this. Structured data is a clarity layer. It helps a crawler resolve that "Dr. [Name], MS Ortho, Tue and Thu 10 to 2, Whitefield" means a named surgeon, at a named facility, on stated days. No platform has published evidence that it gives a ranking bonus, and any agency promising AI placement from schema alone is selling a claim no platform has made.
The four types that carry a hospital
"Why Your Hospital Is Invisible to AI Assistants, and How to Fix It" introduced Physician and MedicalOrganization markup. The vocabulary has moved since, and the move matters for hospitals.
Hospital, on every facility page. Schema.org defines Hospital as both a place and a medical organization, and Google's web index shows it on 10,000 to 100,000 domains (schema.org, Hospital, August 2026 usage band). Each branch gets its own Hospital entity with its own address, telephone, opening hours for OPD and emergency, geo coordinates, and the specialties practised at that site. A network with twelve branches and one Hospital block on the homepage has told the engine it runs one building.
IndividualPhysician, on every consultant page. Schema.org now defines Physician as "an individual physician or a physician's office considered as a MedicalOrganization", with two subtypes: IndividualPhysician for the person and PhysiciansOffice for the practice (schema.org V30.1, 16 September 2026). Plain Physician still appears on 10,000 to 100,000 domains (schema.org, Physician, August 2026 usage band), and on a consultant page it can leave the engine unsure whether it is reading a doctor or a clinic. Use IndividualPhysician, and connect it to the facility with practicesAt, which points at the Hospital entities where that doctor consults. Its usage band is 1,000 to 10,000 domains (schema.org, IndividualPhysician, August 2026), so a hospital that adopts it now is early.
MedicalSpecialty, on both. MedicalSpecialty is an enumeration with fixed values such as Cardiovascular and Oncologic. The list follows broad specialty groupings, and many Indian sub-specialty labels have no exact value. Use the nearest enumeration value in the medicalSpecialty field, and state the precise sub-specialty and the conditions treated in the visible page text, where "The doctor profile is the unit of hospital AEO" already argues it belongs.
FAQPage, where the questions are real. Its status changed this year, covered below.
Two further properties are worth the effort. availableService lets a Hospital or IndividualPhysician list the procedures and tests actually offered, which is the field that answers "does this hospital do knee replacement". hospitalAffiliation records the formal link between a consultant and a hospital. Keep the service list short and exact. A list that includes every test the lab can send out creates the "services not offered" error described in "When the AI gets your hospital wrong".
The one Indian gap to handle carefully: schema.org has a property for the US National Provider Identifier and nothing specific for the National Medical Commission registration. Put the registration number in visible page text and, if you mark it up, use the generic identifier property. No engine has said whether it reads that property, so the visible text is the copy that matters.
Markup has to match the page and the profile
Google's structured data guidelines require that markup is "a true representation of the page content" and forbid marking up content that readers cannot see (Google Search Central, "General structured data guidelines", last updated 10 July 2026). For hospitals this rule does more work than any type choice.
Three mismatches are common on hospital sites.
The OPD hours in the JSON-LD were copied from a template and differ from the hours in the page text. The engine now holds two answers from the same URL.
The branch address in the markup differs from the address on the branch's Google Business Profile by a floor number, a landmark or a PIN code. An engine comparing sources may treat them as two locations.
The consultant's name in the markup uses one spelling, the page heading another, and the Google Business Profile a third. "The doctor profile is the unit of hospital AEO" covers why the doctor record has to agree across four sources. Markup is a fifth copy of that record, and it goes stale on the same day the others do.
The practical rule for a marketing team: generate the markup from the same source record that feeds the page and the profile, and never hand-edit JSON-LD in isolation. If your CMS cannot do that, expect the markup to drift out of step with the page, and stale markup states a wrong fact more cleanly than stale prose does.
Where schema has no effect
Markup cannot fix a wrong fact. It publishes the wrong fact in a more machine-readable form.
Markup cannot outweigh a contradicting source. If three directories list a departed consultant at your hospital, correct JSON-LD on your own site is one voice against three. That is the correction loop in "When the AI gets your hospital wrong", and schema is only one step inside it.
Markup cannot make an engine recommend you. Recommendation depends on what the engine retrieves across the web, and "Share of answer: how to measure whether AI recommends your hospital" sets out how to measure it.
Markup does not make a hospital bookable by an agent on its own. Schema.org's ReserveAction marks where a booking can happen, and "Agents that book: what changes when an AI assistant chooses the hospital" covers what an agent needs beyond that.
How to test it
Use two tools, for two different questions.
The Schema Markup Validator at validator.schema.org checks whether your JSON-LD is valid against the whole schema.org vocabulary, including Hospital, IndividualPhysician and MedicalSpecialty. Run every template through it: facility page, consultant page, specialty page.
Google's Rich Results Test checks only the features Google supports for display. A clean result there tells you nothing about whether your Hospital or IndividualPhysician markup is correct, and since June 2026 it no longer reports on FAQ markup.
Then run the test that matters, a field-by-field comparison. For a sample of consultants and every branch, put the JSON-LD values, the visible page text and the Google Business Profile side by side. Any difference is a defect, whatever the validators say.
Where to start: the Doc Mirror audit
Before anyone rewrites a template, find out what the engines say now. Doc Mirror, the visibility audit Nextdot built, queries ChatGPT, Perplexity, Gemini and Google AI Overviews to see how a doctor or hospital is described, and also audits Google, search results, YouTube and PageSpeed. It also reads and validates the schema.org markup on hospital and doctor pages.
That audit shows which facts the engines state wrongly and where your own markup disagrees with your pages. Fix those records first, then publish markup generated from the corrected source, so that the structured version and the visible version finally say the same thing.
FAQ
What schema should a hospital use? Use Hospital for each facility page, with its own address, hours, telephone and specialties. Use IndividualPhysician for each consultant page, linked to the facilities through practicesAt. Add MedicalSpecialty values to both, availableService for procedures offered, and FAQPage where a page answers real patient questions in visible text.
Does structured data help with ChatGPT? We found no statement from OpenAI, as of September 2026, on whether ChatGPT reads schema markup directly. Google says AI Overviews need no special schema, and Microsoft Bing has said schema helps its language models understand content. Treat markup as a way to make your facts unambiguous, and measure the effect through share of answer rather than assuming it.
What is Physician schema? Physician is the schema.org type for "an individual physician or a physician's office considered as a MedicalOrganization". It now has two subtypes: IndividualPhysician for a person and PhysiciansOffice for a practice. Hospital consultant pages should use IndividualPhysician.
How do you test structured data? Validate every page template in the Schema Markup Validator at validator.schema.org, which covers the full vocabulary. Google's Rich Results Test only covers Google display features. Then compare the markup, the visible page and the Google Business Profile field by field, because a valid block with a wrong fact still fails.