AI Visibility for Medical Practices: What Patients Ask and What AI Gets Wrong

Patients ask AI assistants practical, local questions. Most of them hinge on one fact that is either right or wrong. Here are realistic examples:

1. "Which family doctors in Columbus are accepting new patients?"
2. "Is there a dermatologist near me that takes my insurer's PPO plan?"
3. "Can I see a primary care doctor by telehealth today in Ohio?"
4. "Which doctors at this practice do sports medicine?"
5. "What time does the clinic on Main Street open on Saturday?"
6. "Good endocrinologist for thyroid problems in my city?"
7. "Does this pediatric office do same-day sick visits?"
8. "Is Dr. Patel still at this practice?"

None of these is a request for medical advice. They are booking questions. A wrong answer sends a patient to a competitor or to your front desk with a complaint.

The next move: write down five to eight prompts in your patients' words, including your town and your specialties. The free checker on this site shows whether AI assistants name you for three of them. The same logic applies to dental clinics and mental health practices, with different facts at stake.

Each engine pulls from a different place, so each needs its own check.

• Google AI Overviews and AI Mode. These sit on top of local search. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link, and that there are no additional technical requirements.
• Gemini. A separate Google product. Its Gemini Apps use are governed by Google-Extended, not by AI Overviews rules.
• ChatGPT. ChatGPT search depends on OpenAI's OAI-SearchBot crawler, as of October 2026.
• Microsoft Copilot. Copilot grounds on the Bing index, so Bing Webmaster Tools tells you something Google Search Console cannot.
• Perplexity and Claude. Each runs its own search crawler (PerplexityBot, Claude-SearchBot).

One more constraint: retrieval is per query. When an assistant answers without searching, it relies on training data frozen at a cutoff. A physician who joined after that cutoff can be missing from those answers no matter what you publish. Nobody controls that stage this quarter.

The practical rule: measure each engine on its own. A blended score hides the engine that is wrong about you.

Health is a topic where a wrong answer can hurt someone. Google treats it with extra care, and many vendors turn that into an "E-E-A-T score" to improve. Google says E-E-A-T isn't a specific ranking factor. There is no number to move, and any tool that reports one invented it.

What that means for a practice:

• Name each clinician, with credentials and the specialties they actually practice. Patients want this. It is good practice, not a proven AI lever.
• Keep clinical pages accurate and current because patients read them, not because a date stamp gets cited.
• Skip structured data as an AI tactic. Google says it isn't required for generative AI search, and Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema. Existing markup does no harm, so there is no need to remove it.

The one on-page signal with experimental support is a plain, specific fact on a fetchable page. "Dr. Lee is accepting new adult patients" gives an engine a sentence it can quote. If your roadmap still carries schema or llms.txt tasks, see retiring unsupported GEO tasks.

AI platforms get specific facts wrong about specific practices. For medical offices, the usual suspects are:

• Provider names. A physician who left two years ago still listed, or a new hire missing.
• New-patient status. "Accepting new patients" from an old directory page, after the panel closed.
• Insurance networks. A plan you dropped, or a plan you joined that no page mentions.
• Hours. Saturday hours, lunch closures and holiday schedules that changed.
• Locations. A closed satellite office or a moved suite number.
• Telehealth. Which states you can see patients in, and which visit types qualify.
• Specialties and procedures. A service one provider offered, attributed to the whole practice.

The causes are mostly fixable. Stale third-party listings, contradictions across your own site, and facts that only exist inside a portal are all in your control.

Say you run a hypothetical four-physician family practice. Your about page lists four doctors, your footer lists five, and your insurer's directory lists three. An engine has three answers to choose from. Pick one source of truth per fact, then make every page and listing match it.

When engines cite other sites instead of yours, being named correctly on those sites is the route into the answer. The evidence shows an association, not a guaranteed effect. For practices, check:

• Your Google Business Profile and Bing Places listing, one per location.
• Insurer provider directories for every plan you accept.
• Hospital and health-system affiliation pages that list your physicians.
• Physician directory and review sites where patients search by specialty.
• Your state medical board's license lookup, which carries each provider's legal name.
• Public booking pages run by your scheduling vendor.

Fix departed providers first, then hours, then insurance. Keep a spreadsheet of every listing, its owner and its last review date. For the broader method, see managing your brand as an entity. To see which nearby practices AI recommends instead of you, try a free competitor check. Pharmacies face the same directory sprawl.

Access is where the strongest evidence sits. Run these checks before any content work.

robots.txt, per bot

Each crawler has its own rule. OpenAI states its settings are independent: GPTBot governs training, OAI-SearchBot governs ChatGPT search.

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

Blocking GPTBot is a training choice and does not remove you from ChatGPT search. The details are in should I block GPTBot.

Firewall and CDN

A firewall can return 403 to a crawler that robots.txt allows. Healthcare sites often run strict bot protection. Since 2026-09-15, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and free-tier customers, while Search stays allowed. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.

Portals and JavaScript

Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. A provider list or insurance table rendered by a booking widget may not exist for them. Facts behind a patient-portal login are invisible to every crawler. Put provider names, hours, locations, insurance and new-patient status in the public HTML. More on code-level issues that stop AI indexing.

Some medical content should not be paraphrased by an AI. Examples include patient-specific preparation instructions, medication adjustment guidance and after-hours triage scripts. Google and Bing document on-page controls that keep content out of AI answers.

<meta name="robots" content="nosnippet">
<div data-nosnippet>Pre-procedure instructions...</div>

• nosnippet or max-snippet:0 removes the page's text from snippets and AI features.
• data-nosnippet suppresses one section. Bing says content marked this way is still indexed normally but excluded from snippets and AI summaries.
• noindex removes the page from the index entirely.

These are suppression tools only. No markup makes a section more likely to be used. The common failure is a template that applies them site-wide, hiding your hours page by accident. Audit every template after a redesign, and keep these controls off anything a patient needs to book.

Week 1: inventory. List your providers, locations, hours, insurance plans, telehealth states and new-patient status. Pick one owner for each fact. Write your eight patient prompts.

Week 2: access. Read robots.txt per bot. Test each AI crawler against your firewall. Compare raw HTML with the rendered page for your providers, insurance and hours pages. Move any fact that only appears after JavaScript or behind a login onto a public page.

Week 3: your own site. Remove contradictions between the about page, footer, location pages and provider bios. State each fact in one plain sentence. Check templates for stray nosnippet or data-nosnippet. Apply snippet controls deliberately to instructions that should not be summarized.

Week 4: the outside world. Correct Google Business Profile, Bing Places, insurer directories and affiliation pages. Check index status in Google Search Console and Bing Webmaster Tools separately. Rerun your prompts on each engine and log what each one says.

Then repeat monthly. You cannot force a citation. You can remove every block and every wrong fact in your control.

Should we block AI crawlers to protect patient privacy? Public crawlers only read public pages. Patient data belongs behind authentication, where no crawler reaches it. Blocking search crawlers like OAI-SearchBot removes your public pages from that engine's answers. If you object to training use, block training bots like GPTBot separately, since OpenAI says each setting is independent.

Does adding schema markup help a clinic appear in AI answers? There is no credible evidence that it does. Google says structured data isn't required for generative AI search. Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema. Existing markup does no harm. The better use of time is making sure your facts are public, consistent and in the HTML.

Why does ChatGPT list a doctor who left our practice? The answer may come from training data frozen before the change, or from a stale third-party page it retrieved. You cannot edit training data. You can remove the doctor from every page you control, update directories and affiliation listings, and state your current provider list plainly on a crawlable page.

Can Google Analytics show our AI visibility? Only partly. Some engines tag referral links and others send no referral at all, so analytics sees a fraction of AI-driven visits and none of the answers that never produced a click. See [whether Google Analytics can measure AI visibility](/questions/google-analytics-ai-visibility) for what the reports do and do not show.