AI search visibility for medical practices

Patients now ask AI assistants which dentist, chiropractor or med spa to try, and the answer can include links to the sources the assistant used. Practice owners then get offers to “rank in ChatGPT”. Before paying for that, it helps to know what the companies behind these assistants actually publish about how their systems find, read and cite websites. Each of them documents its crawlers and the controls a site owner has, and those documents are short and specific.

This guide covers the crawlers that OpenAI, Anthropic, Perplexity, Microsoft and Apple publish, how robots.txt and page tags affect search versus model training, what Microsoft says about being cited in Copilot, why a practice’s own facts have to match everywhere, and the FTC rules that apply when someone offers to buy placement or reviews. For Google’s AI features, see our guide to Google AI Overviews and AI Mode. For writing the pages themselves, see our guides to AI-generated content in healthcare marketing and medically reviewed health content. If your clinicians answer questions on public Q&A sites, our guide to Quora for healthcare practices covers disclosure and HIPAA limits.

This is general information, not legal advice. Sources cited were checked on November 1, 2026: OpenAI’s Overview of OpenAI Crawlers (no date shown) and its help article Searching the web with ChatGPT (shown as updated in late October 2026); Anthropic’s help article on its crawlers (April 7, 2026); Perplexity’s crawler documentation (page modified October 29, 2026); Apple’s About Applebot (published September 4, 2026); the Bing Webmaster Guidelines (no date shown); RFC 9309, the Robots Exclusion Protocol (September 2022); 16 CFR 465.2, 465.4, 465.5, 465.6 and 465.7; 16 CFR 255.5(a); 15 USC 45(a)(1); and 45 CFR 164.502(a). These companies change their documentation, so check the current versions. We found no federal rule or agency guidance written for AI search visibility of medical practices as such. State rules differ, so have a health care attorney review your advertising.

The short version

Each company publishes separate crawler names for search, for model training and for fetches a user asks for. Blocking a training crawler does not, by itself, block the search crawler.
OpenAI says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Perplexity and Anthropic say blocking their search crawlers may reduce a site’s visibility in their results.
OpenAI and Perplexity recommend allowing their published IP ranges. A firewall or hosting rule that blocks them can keep a practice out of results even when robots.txt allows them.
Microsoft’s Bing Webmaster Guidelines say they cover Bing search, Copilot and grounding results, and that SEO basics still apply to AI experiences.
OpenAI says placement in ChatGPT search results is not assured, and Microsoft says generative engine optimization does not assure citations. Treat a promise of citations as a sales claim, not a fact.
robots.txt is not a security control. RFC 9309 says its rules are not a form of access authorization, so patient information belongs behind a login, not behind a robots.txt line.
The FTC’s consumer review rule bars buying reviews that express a particular sentiment, fake reviews, and presenting a review site the business controls as independent.

How the assistants find practice websites

An AI assistant that answers with current information has to get pages from somewhere. The companies we checked describe three kinds of automated visitors: a search crawler that builds an index for answers, a training crawler that collects content that may be used to train models, and a user-triggered fetcher that loads a page when someone asks a question. Each kind has its own name, called a user agent token, and robots.txt rules can address each token separately.

OpenAI (ChatGPT): OAI-SearchBot is for search and is used to surface websites in ChatGPT’s search features. GPTBot crawls content that may be used to train OpenAI’s foundation models. ChatGPT-User visits a page when a user’s question calls for it, and OpenAI says robots.txt rules may not apply to it because the action is user initiated. OpenAI says each setting is independent and that search results can take about 24 hours to reflect a robots.txt change.
Anthropic (Claude): ClaudeBot collects content that could contribute to model training. Claude-SearchBot indexes content to improve search results. Claude-User retrieves pages when a user asks Claude a question. Anthropic says its bots honor robots.txt and that blocking Claude-SearchBot or Claude-User may reduce a site’s visibility in user search.
Perplexity: PerplexityBot surfaces and links websites in Perplexity’s search results, and Perplexity says it is not used to crawl content for AI foundation models. Perplexity-User fetches a page when a user asks a question, and Perplexity says this fetcher generally ignores robots.txt rules because a user requested the fetch.
Microsoft (Bing and Copilot): Bingbot crawls for Bing, and the Bing Webmaster Guidelines say Bing and Copilot rely on the same crawling, indexing and ranking foundation. The guidelines list blocking Bingbot in robots.txt among things to avoid.
Apple (Siri, Spotlight and Safari): Applebot crawls for search features in Apple products. Apple says Applebot data may also be used to train its foundation models, and that publishers can opt out of that training by disallowing Applebot-Extended, while Applebot can still crawl for search.

OpenAI’s help article on searching the web with ChatGPT also says ChatGPT search can send rewritten queries to partner search providers, and it lists Microsoft’s privacy statement among them. A practice that is missing from Bing’s index has one more reason to fix that.

Search versus training: what each control does

A practice may want to appear in AI answers while keeping its writing out of model training. The documents we read describe these controls:

OpenAI: allow OAI-SearchBot and disallow GPTBot. OpenAI says disallowing GPTBot indicates the content should not be used in training its foundation models, and that the two settings are independent.
Anthropic: Anthropic says restricting ClaudeBot signals that the site’s future materials should be excluded from its training datasets. Claude-SearchBot is a separate token.
Apple: disallow Applebot-Extended to opt out of training. Apple says a nosnippet meta tag on specific content opts it out of broad world knowledge answers in Siri and Search, and that the content stays discoverable through Spotlight, Siri and Safari.
Microsoft: the Bing guidelines say NOARCHIVE prevents content from being used in Copilot responses and grounding results, NOCACHE limits Copilot to the URL, title and snippet, and NOSNIPPET and DATA-NOSNIPPET may limit citation quality. NOINDEX keeps a URL out of Bing search, Copilot and grounding results.
Google: Google’s controls, including Google-Extended, are covered in our Google AI Overviews guide.

Two details from Apple’s page are easy to miss. If robots.txt does not mention Applebot but does mention Googlebot, Applebot follows the Googlebot rules. And Applebot does not follow crawl-delay. Anthropic says it supports crawl-delay as a non-standard extension.

A blanket rule such as “User-agent: * Disallow: /” blocks each crawler that honors robots.txt, including search crawlers. No federal rule we found requires a practice to allow or block any of these crawlers. We recommend deciding search and training separately, writing a group for each token by name, and checking the file after each website change.

When robots.txt is not the problem

A site can allow each token by name in robots.txt and still be invisible if the server or firewall refuses the requests. OpenAI says to allow OAI-SearchBot and to allow requests from its published IP ranges. Perplexity recommends the same for PerplexityBot and publishes firewall setup steps. Anthropic says blocking its IP addresses may not work as an opt-out because it stops its bots from reading robots.txt. Bot protection on a hosting plan or content delivery network can block these requests without anyone at the practice knowing.

RFC 9309, the Robots Exclusion Protocol, adds two rules that matter here. If robots.txt returns a server error in the 500 range, a crawler must assume complete disallow, so a broken robots.txt can block the whole site. If it returns a status in the 400 range, a crawler may access any resource. RFC 9309 also says crawlers should not use a cached copy for more than 24 hours, unless the file is unreachable.

No federal rule we found sets how a practice checks this. We recommend asking whoever runs the website to look in the server logs for the user agent names above, to confirm robots.txt returns status 200, and to confirm the firewall does not challenge the published IP ranges.

What Microsoft says helps pages get cited

Among the companies we checked, Microsoft’s Bing Webmaster Guidelines give the longest list of what supports being selected for AI answers. They say the guidelines cover Bing search, Copilot and grounding API results, and that practices supporting crawling, indexing and content clarity also support eligibility for grounding results and citations. The points that apply to a practice site:

Help Bing find URLs through XML sitemaps that list only canonical URLs, crawlable internal links and IndexNow notifications when pages are added, updated or removed.
Keep each important URL reachable through standard links, and use 301 redirects for permanent moves.
Do not hide critical content behind client-side rendering.
Make each page focused on one topic, put essential information near the top, and make facts and definitions explicit on the page itself.
Use clear and consistent naming for people, organizations and locations.
Use structured data that reflects visible content. The guidelines say structured data may support clearer grounding but does not assure visibility.
Keep content accurate and up to date, and remove or revise outdated information.

The same guidelines list practices that can reduce visibility or lead to removal: cloaking, link schemes and artificial promotion, scraped content, keyword stuffing and language engineered to trigger citations, automatically generated content at scale without editorial review, misleading structured data, and content designed to manipulate or interfere with the language models used by Bing or Copilot. Hidden text written to instruct an AI assistant falls in that last group.

Your practice facts have to match

When an assistant answers a local question such as “chiropractor open Saturday near me”, it has to decide which practice is which. Bing’s guidelines ask for clear and consistent naming of organizations and locations. Your practice name, address, phone number, hours and providers should read the same on your website, your Google Business Profile, your Apple Maps place card, Bing and the directories you control.

For the details of each listing, see our guides to Google Business Profile for doctors, Apple Maps listings and medical directory listings. For markup that states these facts in a form machines read, see our schema markup guide. Markup has to match what the page shows, a rule the Bing guidelines state and our schema guide covers for Google.

Patient information and AI crawlers

A crawler that honors robots.txt will skip a disallowed path, but RFC 9309 says the rules are not a form of access authorization, and its security section says listing paths in robots.txt makes them discoverable. A user-triggered fetcher may ignore robots.txt, as OpenAI and Perplexity say of theirs. So robots.txt does nothing to protect patient information.

Under HIPAA, a practice may not use or disclose protected health information except as the Privacy Rule permits or requires (45 CFR 164.502(a)). Pages that hold patient information, such as forms, portals and appointment confirmations, belong behind a login, not on public URLs that any crawler can reach. Patient stories and testimonials on public pages are a separate question covered in our testimonials and before-and-after photos guide.

Paying for placement, fake reviews and “best of” lists

Certain offers to get a practice cited by AI assistants rely on reviews or on lists that look independent. The FTC’s rule on consumer reviews and testimonials, 16 CFR Part 465, applies to these tactics:

Section 465.2 prohibits a business from writing, creating, selling or buying reviews that misrepresent that the reviewer exists, that the reviewer had experience with the business, or what that experience was. It also covers procuring such reviews from the business’s own officers, managers, employees or agents, or their immediate relatives, for posting on a third-party site.
Section 465.5 requires clear disclosure when an officer or manager writes a review of the business, and it bars an officer or manager from soliciting reviews from employees, agents or relatives in ways that lead to reviews without a disclosure of the relationship.
Section 465.4 prohibits giving compensation or other incentives in exchange for, or conditioned on, reviews that express a particular sentiment, positive or negative.
Section 465.6 prohibits a business from materially misrepresenting that a website or entity it controls, owns or operates provides independent reviews or opinions about a category of businesses that includes its own. A “best dentists in [city]” site run by one of the dentists falls under this.
Section 465.7 prohibits suppressing reviews through unfounded legal threats, intimidation or false accusations, and prohibits a business that shows reviews on its own website from misrepresenting that the displayed reviews reflect the reviews submitted when it hides reviews based on their rating or negative sentiment.

When a practice pays to be placed in an article or list, the FTC’s Endorsement Guides say a material connection between the endorser and the seller that the audience would not expect must be disclosed clearly and conspicuously (16 CFR 255.5(a)). Section 5 of the FTC Act prohibits unfair or deceptive acts or practices in commerce (15 USC 45(a)(1)). The Bing guidelines separately list link buying and artificial social promotion schemes among manipulative practices.

For how to ask for reviews within these rules, see our patient review replies guide.

Measuring AI visibility without invented numbers

The Bing guidelines say content may appear as impressions, citations or grounding references in Copilot responses without a click, and that a decline in clicks need not indicate a loss of visibility. They suggest monitoring impressions, indexing status, and grounding and citation eligibility.

No federal rule we found governs how a practice measures this. We recommend tracking what you can verify: server log visits from the user agent tokens above, referrals from the assistants’ domains in your analytics, and the answers you get when you ask each assistant about your own practice by name and city on a set date each month. Treat any vendor’s citation share or AI ranking figure as a claim to question.

Questions to ask an agency selling AI search work

Which crawlers does our robots.txt allow and block today, by name, and why?
Does our firewall or host challenge OpenAI’s, Perplexity’s or Anthropic’s published IP ranges?
Does robots.txt return status 200, and what happens if it returns a server error?
Do our name, address, phone number and hours match on our site, Google, Apple Maps, Bing and the directories we control?
Does our structured data match what each page shows?
Will any part of this work involve paid reviews, review incentives, lists we control, or paid placements without disclosure?
Will you add hidden text or instructions aimed at AI assistants? If yes, how does that fit the Bing guidelines on manipulation?
What will you report to us, and which of those numbers can we check ourselves?

Use and avoid

Use: separate robots.txt groups for each search and training token, chosen on purpose.
Use: a firewall rule that lets the search crawlers’ published IP ranges through.
Use: one version of your practice facts, repeated on your site and each listing you control.
Use: pages that answer one patient question each, with the key facts near the top.
Use: structured data that matches the visible page.
Avoid: “User-agent: * Disallow: /” left over from a staging site.
Avoid: relying on robots.txt to keep patient information private.
Avoid: hidden text written to steer AI assistants.
Avoid: paying for reviews, incentives tied to positive reviews, or “independent” review sites you control.
Avoid: contracts that promise citations or AI rankings.

An illustration

This example is invented. [Practice name], an integrative medicine clinic, asks ChatGPT and Perplexity about integrative clinics in [city] and does not see its name. Its web person checks robots.txt and finds a rule left from a site rebuild that blocks all crawlers. They replace it with groups that allow OAI-SearchBot, Claude-SearchBot, PerplexityBot, Bingbot and Applebot, and disallow GPTBot, ClaudeBot and Applebot-Extended, because the owners do not want their articles used for training. The server logs show requests from OpenAI’s crawler getting a challenge page, so the host adds the published IP ranges to the firewall allow list. The office manager finds an old suite number on two directories and fixes them. An agency then offers a package of “[number] AI citations a month” that includes posting reviews from staff family members. The clinic declines, because 16 CFR 465.2 and 465.5 cover reviews procured from staff relatives and insider reviews without disclosure, and because no assistant company we checked promises citations. A month later the clinic asks the same questions on the same date and writes down what it sees.

Checklist

Read your live robots.txt and list each crawler token it allows or blocks.
Decide search and training access separately for OpenAI, Anthropic, Perplexity, Apple and Google.
Confirm robots.txt returns status 200.
Confirm your host or firewall does not block the published IP ranges of the search crawlers you allow.
Submit an XML sitemap of canonical URLs to Bing Webmaster Tools and Google Search Console.
Match your practice name, address, phone number and hours across your site and each listing you control.
Check that structured data matches each page.
Keep patient forms, portals and confirmations behind a login.
Remove any hidden text aimed at AI assistants.
Check review requests and paid placements against 16 CFR Part 465 and 16 CFR 255.5.
Record what each assistant says about your practice on a set date each month.

How we help

We build websites, local and organic search, Google Ads and lead generation and an AI front desk for independent practices. For AI search, that means a robots.txt and firewall setup chosen on purpose, practice facts that match across listings, pages that answer one question clearly, and no tactics that the assistant companies or the FTC treat as manipulation. Our tools are built with HIPAA in mind. We confirm in writing what a BAA covers before any patient information flows through a tool, and if it is not covered, we do not route patient information through it. We are not a healthcare provider and we do not give legal or medical advice. Read our guide to healthcare marketing rules for the wider picture.

How do I get my practice to show up in ChatGPT search?

OpenAI says that to make a website eligible for inclusion, a site should allow OAI-SearchBot to crawl it and confirm that the host or content delivery network allows traffic from OpenAI’s published searchbot IP addresses. OpenAI also says placement is not assured. Beyond that, clear pages and consistent practice facts help each search system, and Microsoft’s Bing guidelines say SEO basics apply to AI experiences.

If I block GPTBot, will my practice disappear from ChatGPT?

OpenAI says OAI-SearchBot and GPTBot are independent settings. Blocking GPTBot indicates your content should not be used to train OpenAI’s foundation models. Search visibility in ChatGPT is controlled through OAI-SearchBot.

Can robots.txt keep patient information away from AI crawlers?

No. RFC 9309 says robots.txt rules are not a form of access authorization, and OpenAI and Perplexity say their user-triggered fetchers may not follow robots.txt. Patient information belongs behind a login, and HIPAA limits its disclosure under 45 CFR 164.502(a).

Does structured data get a practice cited by AI assistants?

Microsoft’s Bing guidelines say structured data may support clearer grounding but does not assure visibility, and that markup must reflect visible content. We found no assistant company that says markup alone leads to citations.

Can we ask staff and friends to post reviews so assistants recommend us?

The FTC’s consumer review rule prohibits a business from procuring reviews from its officers, managers, employees, agents or their immediate relatives that misrepresent the reviewer’s experience (16 CFR 465.2(c)), and it requires disclosure of insider relationships in the cases set out in 16 CFR 465.5. It also prohibits incentives conditioned on reviews of a particular sentiment (16 CFR 465.4).

Sources

More in Local SEO and reviews for practices

Getting a practice cited by ChatGPT, Claude, Perplexity, Copilot and Siri: what each company publishes