ChatGPT and a physio practice: what OpenAI's crawler page settles

What does OpenAI's crawler page settle?

It settles one thing per token it documents. Each token is its own decision in robots.txt, because under RFC 9309 a group of rules written for one crawler does not bind another. The table keeps to what OpenAI states and sets beside each statement what it means for a physio practice's site.

Where no group in the file names OAI-SearchBot or GPTBot, each falls back to the rules written for *. ChatGPT-User cannot be read that way. Because a person starts each of its fetches, OpenAI says robots.txt rules may not apply to it, so do not count on a robots rule to keep it from any page. The crawler access check reports what your file asks of each token it lists, OAI-SearchBot, GPTBot and ChatGPT-User among them; for ChatGPT-User, that is a request OpenAI says may not apply.

What the page leaves open

It does not describe how ChatGPT decides which practice to name, in what order or why, and we do not claim to know. An answer can come from two places, and the page speaks to one of them. With search on, ChatGPT answers from pages its search finds, which for your site means pages OAI-SearchBot was allowed to fetch. Without search, it answers from what the model learned before its knowledge cutoff, which for a practice can mean last year's fees and a therapist who has since left.

Both routes lead back to your own pages. Correct a fee there, and the next fetch finds the new one; leave it, and either route can repeat the old one. Your pages are the part of the picture you write yourself, and the same pages serve Claude, Gemini and Perplexity as well.

The booking system, the intake forms and the public pages

If your practice books online, its site has two halves. One is public: pages about the practice, its treatments, therapists, fees and hours. The other sits behind a login: the booking calendar, intake and health history forms, home exercise programs, messages, invoices. OpenAI's crawlers, like every AI crawler, are for the first half. None should reach the second, and the only thing that keeps it that way is the login.

robots.txt cannot do the login's job. It is a public file, readable by anyone who adds /robots.txt to your domain, so a line disallowing /intake/ tells every reader where the intake forms are. It is a request to crawlers, not a barrier, and OpenAI's page says its rules may not apply to ChatGPT-User at all. If a patient pastes the address of an intake form into ChatGPT, the login is what answers.

The public half needs its facts in text. A booking widget that draws visit types and fees with a script can hand a fetcher that runs no scripts an empty box. Our audit's crawler ran none, and in our audit of 9,483 small business websites (40 trades, 90 metros, late August to early September 2026), 3.0% rendered their content only in JavaScript. Beside the widget, write which visits can be booked online, whether new patients can book that way, and the number to call.

What a practice's public pages carry for ChatGPT search

OpenAI's page describes access, not content, so this part is reasoning: an answer built from search can only repeat a fact that a page it fetched states. The facts a physio patient asks about are the condition and the treatment by name, who provides it and with what license, whether they can book without a referral, the fee and the plans billed, the session length, home or clinic, and the hours. The overview for physiotherapy practices maps each one to a page. For a crawler, each has to be:

• On a public URL of its own: a page for each treatment, each therapist, each location, the fees and the booking policy.
• Linked from the homepage within a click or two. A page reachable only from inside the booking system is not reachable at all, and internal linking is how a crawler that starts at your homepage finds the rest.
• Named and placed: the practice and the town in the title and the first paragraph, so a passage taken from the page still says whose it is.
• Dated where it changes: "Fees as of [month, year]" beside the fees.
• In HTML, not in a PDF fee schedule or a picture of the opening hours.

How a treatment page is built so that a lifted passage stays true is the subject of treatment pages that survive being quoted; what each one says is in a page for each treatment. None of it is specific to ChatGPT, and that is the point: a page a fetcher can read serves every assistant that fetches it.

Checking what OpenAI's crawlers receive, in the order a fetch meets it

These checks settle whether OpenAI's crawlers can reach and read your pages, nothing more. Whether ChatGPT then names the practice is something only a measurement across runs shows, and the panel for it is in what being named means for a physio practice. If your robots file needs rewriting, robots.txt for AI crawlers has one to copy.