AI Search Optimization for Healthcare Websites

Most advice on AI search optimization for healthcare websites starts in the wrong place. It tells a clinic to publish more articles about conditions and treatments, add some markup, and wait for ChatGPT or Google's AI answers to start quoting it. For most clinics, that plan aims at the one kind of question they are least likely to win.

Health is the topic where AI search engines are most careful about what they say and who they say it from. When a patient asks a general medical question, the answer tends to lean on national health bodies, large hospital systems and medical publishers. A three-doctor practice is rarely going to be the source for "what causes lower back pain."

But patients ask AI a second kind of question, and on that one a clinic's own website is often the best source that exists: "Is this clinic taking new patients?" "Do they accept my insurance?" "Which doctor would I actually see, and are they properly qualified?" Those questions have local, specific, checkable answers, and nobody can answer them better than the clinic itself. This guide is about winning that second kind of question: what AI engines are careful about in health, which facts a clinic site needs to state plainly, how to make your doctors verifiable, and how to keep all of it accurate over time.

Health Is Where AI Engines Are Most Careful

Google has said this directly. After AI Overviews made some widely shared mistakes in its first weeks, Google's head of Search, Elizabeth Reid, wrote that "for topics like news and health, we already have strong guardrails in place," and that "in the case of health, we launched additional triggering refinements to enhance our quality protections." In plain terms: for health searches, Google is pickier about when an AI answer appears at all, and about what it draws on.

That caution keeps tightening. In January 2026, after a Guardian investigation found AI Overviews giving simplified liver test ranges that ignored age, sex and other factors, Google removed AI Overviews from some of those searches. A Google spokesperson told reporters the company does not "comment on individual removals within Search," but works to "make broad improvements." Variations of the same question could still produce an AI summary, so the removals were narrow, but the direction is clear.

This sits on an older Google rule: its systems "give even more weight to content that aligns with strong E-E-A-T" (experience, expertise, authoritativeness and trust) on what it calls "Your Money or Your Life" topics, and health is the clearest of them.

None of this means a clinic should give up on AI search. It means a clinic should be honest about which questions it can realistically win, and put its effort there.

The Two Kinds of Questions Patients Ask AI

Think about what a patient actually types into ChatGPT, Gemini, Perplexity or Google's AI Mode when a health need comes up. The questions fall into two groups.

Medical knowledge questions. "What are the early signs of gum disease?" "Is it normal for a knee to click after surgery?" "How long does a tension headache last?" These are general. The answer is the same whichever clinic you go to, and the strongest sources are the ones with the most medical authority behind them: national health services, large academic hospitals, established medical publishers. This is exactly where the caution described above applies most heavily. A small practice's article on the same topic is competing against institutions with hundreds of reviewing clinicians.

Provider questions. "Is there a pediatric dentist in Leeds taking new NHS patients?" "Does Riverside Physio accept my insurance?" "Which dermatologist at Oak Street Clinic does mole checks?" "Can I book a same-day appointment at the walk-in clinic on Market Road?" These are specific to one place and one set of people. No national health body knows whether your clinic is accepting new patients this month. The only source that can answer is you, plus a handful of directories that copy (often badly) what you publish.

The second group is where a clinic's website does real work in AI search. When an AI engine has to answer a provider question, it needs facts it can state with confidence: a yes or no, a list, a name, a qualification. If your site states those facts clearly, you are the obvious source. If it doesn't, the engine falls back on a directory listing, an old review, or nothing at all.

There is also a quieter benefit. A clinic whose practical facts are clear and consistent is easier for an AI engine to recommend when someone asks a broader local question. Getting onto the shortlist for "near me" style questions depends heavily on maps and directory data, which we cover in our guide to near me searches. What gets said about you once you are on that list depends on what you have published.

The Practical Questions Only Your Site Can Answer

Here are the provider questions patients ask most often, and where the answers tend to hide on real clinic websites.

"Are you taking new patients?"

This is one of the first things a patient needs to know, and one of the facts clinic websites most often leave out entirely. When it does appear, it is usually in a news-style post from two years ago ("We're pleased to announce we're welcoming new patients!") that nobody has updated since.

What to do: state it on the main location or contact page in a plain sentence, with the date it was last confirmed if it changes often. "We are currently accepting new patients for general dentistry. Our orthodontics list is full until March." A machine can only repeat a fact if the fact is written down somewhere it can read.

schema.org has a dedicated property for this, isAcceptingNewPatients, defined on MedicalOrganization as "whether the provider is accepting new patients." It takes a simple true or false. It is real vocabulary and worth adding, but treat it as a second copy of a fact that already lives in your visible text, not a replacement for it. No AI engine has said publicly that it reads this specific property, and markup that says something the page doesn't is exactly the kind of mismatch to avoid.

"Do you take my insurance?"

In the US especially, this decides whether a patient books at all. On many clinic sites, the answer is a scanned PDF of insurance logos, an image of a list, or a line saying "please call to check your coverage." A PDF can sometimes be read. A picture of logos mostly can't. "Please call" can't be repeated as an answer by anyone.

What to do: publish the plans or networks you accept as plain text on a normal web page. Name them the way patients and insurers name them. If the list changes, change the page, and say when it was last updated. If coverage genuinely varies case by case, say exactly that and say what does vary, which is still more useful than nothing.

The same schema.org type has a matching property, healthPlanNetworkId, described as the "name or unique ID of network" (networks are often shared across several insurance plans). Again: useful as a structured copy, never as the only copy.

"Who would I actually see?"

Patients increasingly ask about a specific doctor, therapist or dentist, not just the clinic. A team page with first names and smiling photos doesn't answer that. A team page where each clinician has their full name, their role, their qualifications, what they treat, which locations they work at and which days, does.

What to do: give each clinician their own page, or at least their own clearly separated section, with the facts in text. Avoid putting qualifications only inside a photo caption graphic, and avoid lumping everyone into one paragraph of prose. The next section goes deeper on this, because for health sites, clinician identity is where trust is decided.

"How do I book, and how soon?"

Many clinics hand booking to a third-party system shown inside a box on their page. That box is often a separate page loaded from someone else's server, and what it shows is usually not read as part of your own page's text. So a patient asking "does this clinic do Saturday appointments?" may get no answer, even though the booking widget shows Saturday slots.

What to do: state the basics in your own words outside the widget. Opening hours, whether you offer same-day or emergency appointments, whether you offer video consultations, whether a referral is needed, and the phone number. The widget handles the booking. Your text handles the questions.

Proving Your Doctors Are Real People With Real Licenses

For a health site, the most important trust question an AI engine can ask is simple: are the people behind this clinic real, and are they qualified to do what the site says they do? A website can claim anything. What makes a claim believable is that it matches something independent.

Point to records you don't control. In the US, healthcare providers who bill insurance carry a National Provider Identifier (NPI), a 10-digit number listed in the NPI Registry, a free public directory run by the Centers for Medicare and Medicaid Services. The registry shows the provider's name, specialty and practice address. State licensing boards publish license lookups, and specialty boards publish certification lookups. In the UK, the General Medical Council, the General Dental Council and the Health and Care Professions Council all run public registers. Most countries have something similar.

When your clinician page states a credential, it should be the same credential, under the same name, that these registers show. A doctor listed as "Dr Sam Patel" on your site, "Samir Patel" on the register and "S. Patel MD" on a review site is three slightly different people to a machine trying to match them up. Pick the registered form of the name and use it everywhere.

Use the right structured data, because the obvious choice is wrong. This is where many health-site markup guides slip, because the type's name sounds like a person. schema.org's Physician type is not a person. Its definition is "an individual physician or a physician's office considered as a MedicalOrganization," and it sits under Organization and LocalBusiness, not under Person. That means person-only properties such as alumniOf (where someone studied) don't belong on it.

schema.org now handles this more cleanly. Physician has two subtypes: IndividualPhysician, defined as "an individual medical practitioner," and PhysiciansOffice. IndividualPhysician comes with a practicesAt property linking the practitioner to the clinic or hospital where they work. And Physician itself now carries usNPI, the 10-digit National Provider Identifier, plus medicalSpecialty and hospitalAffiliation.

A sensible pattern for a clinic site:

• The clinic as the most specific accurate type (MedicalClinic, Dentist, PhysiciansOffice and so on), with isAcceptingNewPatients and healthPlanNetworkId where they apply.
• Each practitioner as a practitioner (IndividualPhysician for a doctor), with medicalSpecialty, usNPI if you are in the US, and practicesAt pointing to the clinic.
• Each clinician as a person (Person) wherever they appear as an author or medical reviewer of content, with their qualifications and sameAs links to the registers and professional profiles that confirm them. Our guide to author schema covers building that person record in depth.

Whatever markup you use, the same rule applies as everywhere else: every fact in the markup must also be visible on the page, and must match the independent records.

What Your Health Articles Are Actually For

If a clinic is unlikely to be the AI source for general medical questions, should it stop writing patient education content? No. But it helps to be clear about what that content does.

It shows who stands behind the clinic. An article on recovering from a knee replacement, written or reviewed by your named orthopedic physiotherapist, tells both readers and machines what your team knows and treats. That supports the provider questions, even if the article itself is never quoted for the general one. Google's own helpful-content questions ask whether it is "self-evident to your visitors who authored your content" and whether bylines "lead to further information about the author." For a clinic, that further information is the clinician page described above.

It answers the local version of a general question. "What to expect at your first physio appointment" is general. "What to expect at your first appointment with us: how long it takes, what to wear, whether you need a referral, and where to park" is yours alone. The second kind is where a clinic's article can be the best available source.

It needs a visible review process. For health content, say who reviewed each page and when, in the text, and keep it true. Our dental and medical schema guide covers the reviewedBy and lastReviewed markup for this; the point here is that the review has to really happen. A "last reviewed" date that never changes is worse than none, because it tells a careful reader the page has been abandoned.

Running the Audit, Fix and Monitor Loop on a Clinic Site

AI readiness for a clinic follows the same loop as any site, covered step by step in our guide to the AI readiness loop: audit, prioritize, fix, apply, monitor. What changes for healthcare is which findings matter most and what tends to break.

Audit with the provider questions in mind. An AI readiness audit checks whether AI crawlers can reach your pages, whether the content is readable as text, whether your business and its people are clearly identified, and whether the site can answer real questions about you without outside help. That last test, which we call answering from the site alone, matters a great deal for clinics, because the practical facts patients want are exactly the ones an engine cannot reliably find anywhere else. Our explainer on the Answer pillar covers how that test works.

Prioritize the facts patients act on. On a clinic site, a missing "accepting new patients" statement or an insurance list stuck in an image usually matters more than a missing markup property on a blog article. Fix what changes a booking decision first.

Expect clinic-specific drift. Clinics change in ways most businesses don't, and each change quietly breaks the facts above:

• A clinician leaves, and their page, markup and booking profile stay up for months.
• A new clinician joins and appears only in a social media post.
• The clinic joins or leaves an insurance network, and the list doesn't change.
• Hours change for a season, and three different pages show three different sets of hours.
• The new-patient list closes, and the "welcoming new patients" banner stays.

Each of these is a moment to re-check, not just a once-a-year audit.

Monitor with practical questions, not vanity ones. When you check what AI engines say about you, ask the questions patients ask: "Is [clinic] taking new patients?", "Does [clinic] accept [plan]?", "Who are the physiotherapists at [clinic]?" A wrong answer to one of those is a concrete, fixable problem. How to run those checks fairly, without your own account history skewing the result, is covered in our ChatGPT check guide. AI Schema Gen can track ChatGPT's answers to a set of questions like these over time; other engines still need checking by hand.

A Question-by-Question Check

Here is the whole idea as a table you can work through. For each question, find where your site answers it today.

Patient question	Where the answer should live	What usually goes wrong
Are you taking new patients?	A dated sentence on the location or contact page	Missing, or an old announcement post
Do you accept my insurance?	A plain-text list on its own page	A PDF, an image of logos, or "please call"
Who are your clinicians?	One page or section per clinician, in text	First names only, credentials inside images
Is this doctor qualified?	Full registered name, qualifications, links to public registers	Name spelled differently than on the register
What do you treat?	A services page naming conditions and treatments in plain words	Vague marketing language ("whole-body wellness")
When are you open, and can I book today?	Hours and same-day policy in text, outside the booking widget	Only visible inside the third-party widget
Do I need a referral?	Stated on the services or booking page	Not mentioned anywhere
Where are you, and is there parking or step-free access?	Location page with address and access details	Only on the map embed

If you can point to a clear, current, text answer for every row, your site is in good shape for provider questions. Every blank row is a question an AI engine will answer from somewhere else, or not at all.

Common Mistakes

Writing for the general question you can't win. Publishing a twentieth article on "signs of a vitamin D deficiency" while the insurance list sits in a PDF puts effort where it is least likely to pay off.

Keeping key facts inside images and widgets. Insurance logos, credential badges and booking-widget hours are all hard or impossible for a machine to read. Put each fact in your own text as well.

Marking up a doctor as a Physician with person details. Physician is an organization type. Use IndividualPhysician for the practitioner and Person for the human as author or reviewer.

Clinician names that don't match the registers. Use the same registered name and qualifications on your site, in your markup, on directories and on review sites.

Treating a structured property as the answer. Adding isAcceptingNewPatients: true while the page itself says nothing about new patients leaves the most widely read copy of the fact, your visible text, empty. Write the sentence first; the markup repeats it.

Announcing changes only on social media. A new clinician introduced in an Instagram post, or a closed new-patient list mentioned only on Facebook, never reaches the pages AI engines read about you. Every change belongs on the website first.

Putting patient details anywhere near your markup. Testimonials that mention a patient's condition raise privacy questions in most countries. Keep them out of structured data entirely, and check anything you publish with whoever handles compliance for your practice.

How to Check Your Clinic's Site This Week

1. Work through the table above. For each row, write down the page that answers it, or "none."
2. Fix the "none" rows first, starting with new patients and insurance. Plain sentences on normal pages.
3. Match every clinician to a public register. Same name, same qualifications. Fix whichever side is wrong.
4. Look at your markup. If your doctors are marked up as Physician with person details, plan the move to IndividualPhysician plus Person.
5. Ask an AI engine three provider questions about your own clinic, in a clean session, and note the answers. Repeat after your fixes go live.
6. Run a free AI readiness check on your main location page to see whether AI crawlers can reach it and whether it can answer questions about you on its own.
7. Put the drift triggers in your diary. Every time a clinician joins or leaves, a network changes, or hours change, re-check the affected pages the same week.