How to Rank in ChatGPT Search: Getting Your Clinic Named When Patients Ask AI

"Rank" is the wrong verb, and the sooner a marketing team stops using it, the faster this gets useful.

There is no position three in ChatGPT. There is an answer, written fresh each time, that either names your clinic or does not. Two patients asking the same question ninety seconds apart can get different answers. Nothing about that behaves like a results page, and any vendor selling you a "ChatGPT rank tracker" is selling you a sample, not a ranking.

What you can influence is whether your page gets retrieved when the assistant goes looking, and whether it is quotable enough to survive into the answer. Those are two separate problems with two separate fixes.

01How the answer actually gets built

When a question needs current information, the assistant does not recall your clinic from training. It runs a search, pulls a handful of pages, and writes an answer from what those pages say, usually with links back.

So there are exactly two ways to be absent. Your page was never retrieved, or it was retrieved and nothing in it was usable.

Almost all of the practical work sits in those two buckets, and the first one is more often the culprit than anyone expects.

02Check that you are not blocking the crawler

This is the single most common reason a clinic is invisible in assistant answers, and it takes ten minutes to check.

OpenAI operates separate user agents for separate jobs. GPTBot is the one associated with training. OAI-SearchBot is the one that builds the search index behind ChatGPT's browsing. ChatGPT-User is the fetch that happens when a user's question triggers a live page load.

A site that added a blanket "Disallow" for GPTBot during the 2023 wave of AI-blocking has made a training decision. A site that blocks OAI-SearchBot has removed itself from the search results. Those are different choices and a lot of practices made the second one by accident, usually by pasting a blocklist from a forum.

Then check the layer above robots.txt. Cloudflare and most WAFs now ship bot-management rules that challenge or block AI crawlers by default, and those rules override whatever your robots.txt politely says. Look at the bot settings, not just the file.

While you are there, confirm your pages render without JavaScript. Content that only appears after a client-side fetch is frequently not seen at all.

03Write so a sentence can be lifted out

A retrieved page still has to be quotable, and medical marketing copy is often written in exactly the style that cannot be quoted.

Three habits fix most of it.

Answer first. Put the direct answer in the first two sentences under the heading, then explain. A page that builds to its conclusion in paragraph nine gets summarised from paragraph one.

Make headings question-shaped, in the words a patient would use. "How long does recovery from a knee replacement take" beats "Post-operative protocol".

Write self-contained sentences. A model lifting one sentence out of your page takes no context with it. "It usually takes six weeks" is useless on its own; "most patients return to desk work about six weeks after a knee replacement" survives the trip.

There is a safety version of this that matters more in healthcare than anywhere else. If a qualifier lives in a separate paragraph from the claim it qualifies, summarisation will drop it. Write the caveat into the sentence — "most cases are benign, but any new lump should be examined" — rather than reassuring in one paragraph and warning in the next.

04Be the same entity everywhere

Assistants resolve entities before they answer. If your clinic appears as three slightly different names across your site, your Google Business Profile, the hospital directory and your consultants' bios, you have created three weak entities instead of one strong one.

Pick the legal or trading name, use it identically everywhere, and do the same for each consultant. Same spelling, same qualifications, same registration number, same address format. It is unexciting and it is the cheapest win available.

The corollary is that third-party corroboration carries real weight here. You are named more readily when several independent sources say the same thing about you — an association member page, a hospital's consultant listing, a genuine press mention, an accurate directory entry. This is not link buying. It is making sure the facts about your practice exist in more than one place and agree with each other.

05The local question is answered from other people's pages

Ask an assistant for the best cardiologist in a given city and watch where the answer comes from. It is rarely a clinic website. It is directories, hospital listings, and "top ten" articles written by publishers.

Which means the play for local visibility has two halves. Be accurate and present on the sources that actually get quoted for your city and specialty, and be the site that answers the underlying clinical question so well that it gets pulled in for the substance even when the recommendation comes from elsewhere.

The second half is the durable one. It is also the same work that earns organic visibility, which is why treating this as a separate discipline from SEO tends to produce a lot of activity and very little movement. The specialised part — entity work, citation building, the prompt panel — sits on top of that foundation rather than replacing it, and it is what our ChatGPT ranking work actually consists of.

06What patients actually ask, and which pages answer it

The questions people bring to an assistant are not the questions they type into a search box. They are longer, more personal, and often embarrassing — which is precisely why the volume is growing in healthcare faster than in most categories.

Four shapes account for most of it.

Symptom triage. "I have had a dull ache in my lower back for three weeks and it is worse in the morning — should I see someone?" The pages that get pulled in here are symptom pages written in patient language, not condition pages written in clinical language.

Comparison. "What is the difference between a stent and a bypass?" Comparison content is unusually well suited to being quoted, because it is already structured as a set of discrete, liftable statements.

Cost and process. "How much does an MRI cost in Hyderabad and do I need a referral?" Most clinic sites refuse to answer this, so the answer comes from an aggregator, and the aggregator gets the citation.

Preparation and aftercare. "What can I eat before an endoscopy?" Low glamour, high volume, almost never written well, and the easiest place to get cited.

A practice that publishes proper pages for those four shapes will get named far more often than one that publishes a blog about industry trends. The E-E-A-T and authority work matters, but it matters on pages that answer something.

07What carries over, and what does not

Carries over: named clinician authorship with real credentials, a visible review date, an actual address and phone number, schema that matches the visible page, and content depth on a defined subject area. This is the same E-E-A-T requirement search has applied to health content for years, now with a second consumer reading it.

Does not carry over: keyword density, thin location pages, and anything written to game a ranking that does not exist.

Contested: The llms.txt file. No major assistant has confirmed it as a retrieval input. It costs very little to publish, so treat it as cheap insurance rather than a strategy, and do not let anyone charge you for it as a deliverable.

Also contested, and worth saying plainly: schema markup on its own does not get you named. It helps a machine parse a page it has already decided to read. It does not get the page read.

08How to measure something with no rank

You cannot pull a position. You can do three things that are genuinely informative.

Watch referral traffic. Sessions arriving from chatgpt.com, perplexity.ai and similar appear in analytics like any other referrer. Low volume, high intent, and the trend matters more than the number.

Run a prompt panel. Write thirty questions a real patient would ask — symptoms, procedures, costs, "best in city" — and run them monthly, logged in a spreadsheet, recording whether you were named and which source was cited. Keep the question set fixed so the months are comparable. This is manual, slightly tedious, and the only measurement here you can fully trust.

Monitor brand mentions. If you are being named without a link, referral analytics will never show it.

Be sceptical of dashboards offering an "AI visibility score". They are running a sample of prompts and extrapolating, which is exactly what your own panel does, without telling you the question set.

09What to do first

1. 1Check robots.txt and your firewall's bot rules for OAI-SearchBot. Today.
2. 2Confirm your key pages render without JavaScript.
3. 3Rewrite your top ten pages answer-first, with question-shaped headings.
4. 4Standardise the clinic name and every consultant's details across every property.
5. 5Put a real named clinician and a review date on every clinical page.
6. 6Build the thirty-question panel and run it once a month.

The full method, including the page templates, is in our AI search optimisation guide.

---

If you want to know whether you are currently blocking the crawlers, how your pages read to a summariser, and where you stand on a fixed panel of patient questions, get a free audit and we will send the findings whether or not you work with us. Or book a strategy call.