We audited 30 medical practice websites. 2 labeled their FAQs for search engines.

What did we find on 30 medical practice websites?

The 30 practice websites we read were easy for AI crawlers to reach and hard for them to read as a business. All 30 sites served their main text in the page itself, with a median of about 990 words on the homepage. That matters because some AI fetchers do not run scripts. Anthropic's documentation for Claude says: "The web fetch tool currently does not support websites dynamically rendered with JavaScript."

The weak spots are the machine-readable layer and the questions. 10 of 30 sites show an FAQ, but only 2 of 30 mark one up with FAQPage schema. 28 of 30 show a phone number on the homepage, but only 6 of 30 put a phone number in business schema.

Can AI crawlers reach medical practice websites?

Yes: none of the 28 robots.txt files we could read blocks an AI search crawler. We tested each file against Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot and PerplexityBot. 2 of the 30 sites served no robots.txt file, which also leaves every crawler allowed. OpenAI describes its search crawler this way: "OAI-SearchBot is used to surface websites in search results in ChatGPT's search features."

Few practices have taken a side. Only 1 of 28 files names any AI crawler at all, to allow it or to block it, and 0 of 28 block a training crawler such as GPTBot or Google-Extended. To read more on what earns a mention, see how a clinic gets named in ChatGPT.

robots.txt is only half of access. Google's AI features guide lists "Ensuring that crawling is allowed in robots.txt, and by any CDN or hosting infrastructure" as a step. A firewall can refuse a crawler the file allows. We saw 1 practice site refuse our ordinary request, and 3 more we could not read at all; we did not count those as blocking AI crawlers, because we cannot tell how they treat a verified crawler.

9 of 30 sites publish an llms.txt file, a plain summary written for AI tools. Google says these files do nothing for its results: "Doing so will neither harm nor help your site's visibility or rankings in Google Search, as Google Search ignores them." Other tools may read them, so we report the count and leave it out of the score. Our page on generative engine optimization covers where the effort pays instead.

Do practice websites tell AI engines who they are?

Most start the job, and few finish it. 26 of 30 homepages carry JSON-LD structured data, and 19 of 30 describe the practice as an organization, a local business or a medical type. 11 of 30 use a medical type such as MedicalClinic or Physician.

The contact details are where it thins out. 22 of 30 homepages show both a phone number and a street address, or a state and ZIP code. Only 6 of 30 put a phone number in their business markup, and 10 of 30 put an address there. Where a site did put a phone number in markup, it matched the page: 6 of 6.

So markup is not a ticket into AI answers. We treat it as hygiene: one clear statement of the practice's name, phone and address that matches the page and the Google Business Profile.

Do practice websites answer patient questions?

Most answer early, and few answer in questions. 20 of 30 homepages open with a full paragraph of 20 or more words inside the first 100 words of main text. 7 of 30 use a question as a section heading on the homepage.

10 of 30 sites have an FAQ section or an FAQ page. 9 of those 10 carry no FAQPage markup, so the questions are written for people and unlabeled for machines. For the bigger picture, read why competitors show up in ChatGPT.

Do practice articles show who wrote them and when?

Most articles we reached carry a date, and few show a clinical review. 19 of 30 homepages link a blog or news section. From 16 of those 19 we reached an article by plain links.

13 of 16 articles show a date, 11 of 16 name an author, and 10 of 16 show both. 2 of 16 carry a "reviewed by" line. We look for a named author and a date because a reader, and an engine quoting the page, both need to know who stands behind a health article.

How did the 30 sites score out of 10?

The median medical practice site passed 6 of the 10 checks. Scores ran from 3 to 9. 5 of 30 sites passed 8 or more, 13 of 30 passed 5 or fewer, and 0 of 30 passed all 10.

What should a medical practice fix first?

Start with markup for facts the site already shows. These five steps follow the gaps in the data, in order of effort.

1. 1Mark up the questions you already answer9 of the 10 sites with an FAQ have no FAQPage markup. The answers exist; the label is missing.
2. 2Put your name, phone and address in schemaUse the same phone number and address the page shows, and the same ones your Business Profile shows.
3. 3Sign and date every articleName the author and the clinician who reviewed it, with a real date. Never invent a reviewer.
4. 4Choose your AI crawler rules on purposeRead your robots.txt file, then ask your host whether bot protection blocks crawlers the file allows.
5. 5Answer the main question near the topSay who you treat, what you offer and where, in one plain paragraph under the main heading.

None of this makes an AI engine name a practice. It removes reasons for an engine to pass a practice by. PELORA offers AI search optimization and healthcare marketing to practices that want help with it. A practice that also runs ads has a separate checklist: health ad rules for clinics.

How did we run this study?

• •

  Sample. We searched 30 phrases: family medicine, gastroenterology and orthopedic practices in 10 large US metro areas (Seattle, Phoenix, Denver, Dallas, Kansas City, Minneapolis, Nashville, Atlanta, Columbus and Boston). For each phrase we took the first result that was a practice's or physician group's own website.

• •

  What we skipped. Hospitals and health systems, academic and residency clinics, insurer-owned groups, directories, job boards and news. Where a phrase returned no practice site, we ran two backup wordings we chose in advance.

• •

  Search tool. The searches ran through our web search tool on October 24, 2026, with directory and job sites filtered out. These are search-tool results, not Google's page one, and rankings change daily.

• •

  Reading the sites. One plain request per page with an ordinary browser user agent: the homepage, robots.txt, llms.txt, a linked FAQ page and one linked blog article. No sign-ins, no forms, no retries.

• •

  Sites we could not read. 1 site refused our request, and 3 could not be read (connection resets and a timeout). We recorded each one, did not retry it, and took the next practice result for that phrase. We tried 34 sites to measure 30.

• •

  Exclusions. We left out PELORA's clients and businesses tied to PELORA's founder. 0 sites in this sample needed to be dropped for that reason.

• •

  Privacy. This page reports totals only and names no practice. The site list stays in our audit file. We collected no patient information.

• •

  Limits. This study measures signals on the sites, not answers from ChatGPT, Perplexity, Gemini or Google's AI features, and we queried no engine. Each check is a script reading of public HTML on one day. Call tracking can swap phone numbers, and our server response times come from a cloud server, so they are lab readings. On that point, web.dev says: "Good TTFB values are 0.8 seconds or less."

Why did PELORA run this study?

What do practice owners ask most about AI visibility?