Why doesn’t ChatGPT recommend my practice? (2026 guide)

A dentist I worked with last year has a website most practices would envy. Photography, scheduling, a page for every service. He asked ChatGPT who the best implant dentist in his city was, and it named four practices. None of them was his. Two of them have websites that look like 2014.

What happens between the question and the answer

Two different things, depending on the question. A "who should I see for X in Y" question usually triggers a live web search, so the gate is whether the engine can crawl and find you right now. A "what is this practice" question is often answered from what the model already absorbed, so the gate is how often you were mentioned across the web over the last few years.

The first kind you can influence in weeks. The second takes months and is built out of other people’s pages: directories, association listings, local press, podcasts, forum threads. Both matter, and a plan that only works on one of them will underdeliver on the other.

The four engines do not read the same web

Each assistant runs its own crawler and its own retrieval. Blocking one bot, at the firewall or in robots.txt, removes you from that assistant and no other. The table below uses the crawler names the vendors themselves document, because those are the names your host or CDN is actually filtering on.

The training crawlers are a separate decision. OpenAI documents GPTBot as the crawler for model training and OAI-SearchBot as the one behind search results, so blocking GPTBot does not remove you from ChatGPT’s answers. Blocking OAI-SearchBot does.

Ranking on Google is a different job from being cited

Ahrefs asked four assistants 15,000 long-tail questions and compared what they cited against Google and Bing results. About 12% of cited URLs also sat in Google’s top ten, and for ChatGPT the URL-level overlap was 8%. That is a vendor study, not peer-reviewed, but the direction matches what practices report.

Read that number the right way. It does not mean search rankings are worthless. It means the page that ranks and the page that gets quoted are often not the same page, so a practice can be first on Google for implants and absent from the answer a patient actually reads.

Most of what moves the answer is not on your website

Assistants cross-check. When your address, phone number, credentials and service list agree across your site, Google Business Profile, Healthgrades, Zocdoc, your state licence board and your association listings, you are one clear entity. When they disagree, you are three half-entities, and the engine names a competitor it can describe with confidence.

This is also where the patient-safety problem sits. In the rater8 survey, 66% of the patients who had used AI to research a provider had run into wrong information: addresses, phone numbers, insurance, hours. 60% of patients trusted the summary without verifying it anywhere else. Stale directory entries are not a marketing problem at that point. They are a booking problem.

What the peer-reviewed evidence says actually works

The strongest study on this is Aggarwal et al., presented at KDD 2024, which tested content changes across roughly 10,000 queries. Adding real statistics, direct quotations and cited authoritative sources raised visibility in generative answers by roughly 30 to 40%. Keyword stuffing produced nothing.

That is a useful instruction for a practice page: put the number in, say where it came from, quote the person who said it, and date it. A page that reads like a brochure gives an engine nothing to lift. A page that says "we see 40 implant cases a year and our failure rate is X, measured over five years" gives it a sentence it can quote and attribute.

What to check this week

1. Ask each of the four assistants the question a patient would ask, logged out, and write down every practice it names. Keep the wording fixed so you can compare next month.
2. Open your robots.txt and ask your host whether the CDN or firewall is blocking OAI-SearchBot, PerplexityBot or Claude-SearchBot. Bot protection blocks these by default more often than people expect.
3. Confirm the site is submitted in Bing Webmaster Tools and Google Search Console. Bing indexation is a gate for more than Bing.
4. Put your name, address, phone, credentials and services side by side across your site, Google Business Profile, and every directory you appear in. Fix the disagreements before writing anything new.
5. Pick the five questions you most want to be named for and check whether any page on your site answers them in the first paragraph rather than the fifth.

Where this does not apply

If your site is new, or was recently rebuilt on a new domain, none of this moves until the site is indexed at all, which takes weeks. If you are hospital-employed, the entity the engines describe is usually the hospital, and your leverage is the profile they publish about you. If you have multiple locations under one brand, fix the location pages and the Business Profiles first: an engine that cannot tell your three offices apart will describe none of them well.

One more limit worth saying plainly. Vendor research through 2025 and 2026 keeps finding that how established a domain already is predicts citation better than any single on-page change, and that 40 to 60% of cited sources change from month to month. So judge this over three to six months and a fixed set of questions, never on one answer on one afternoon.