Talsara

Blog · 31 October 2026 · 8 min read · By Pogos Kazarian

Why doesn't ChatGPT recommend my business? (And how to check)

What ChatGPT reads before it names a local business, five patterns we keep finding when a business is missing, how to check your own, and what to fix first.

The short answer: when someone asks ChatGPT for a local business, it searches the web, reads a handful of pages and names the businesses those pages make easy to recommend for that exact question. If you are missing, it is usually one of five things: your pages don't answer the question that was asked, the sources it reads for your trade don't include you, it has more specific evidence about someone else, your site is hard for its search crawler to read, or you simply weren't in that sample of answers. Nobody outside OpenAI can prove why one business was chosen over another. You can, though, see what the answers cited and what those pages have that yours don't.

What happens when someone asks

Ask ChatGPT "best dental implant clinic in Tulsa" and, with web search on, it does roughly what a careful person would: runs a few searches, opens some of the results, and writes an answer with links to the pages it relied on. The answer is built from those pages, not from a ranked list somewhere.

That makes the links worth more than the answer itself. In our first measurement, 60 answers to 12 patient questions about dental implants in Tulsa, Oklahoma (26 October 2026, gpt-5.4 with web search, location set to Tulsa), the most-cited pages were two practices' own websites, each linked in about 3 of every 10 answers. More than half of the answers (35 of 60) also linked to a directory, a review site or a professional association. The full numbers are in the Tulsa case study.

The spread was wide. Of 51 practices in the Tulsa area that offer implants, 22 were named at least once and 29 never were. The leader was named in 26 of the 60 answers.

1. Your page doesn't answer the question that was asked

This is the pattern we see most. A business offers the service, but its page about it is general, short, or written for a different question.

Two Tulsa practices both offer full-arch implants (All-on-4®, a Nobel Biocare trademark). On five questions (the best implant clinic, affordable implants, All-on-4, same-day implants and full-mouth implants), one of them was named in 0 of 25 answers. Its page on the treatment has about 600 words, mentions the treatment by name once and reads like a general page on rebuilding a smile. The other practice's page has about 1,100 words, names the treatment 20 times, explains who it suits and answers common questions. That practice was named in 3 of 5 answers on two of those questions, and its page was among the cited sources.

We can't prove the page is the reason. But when the answers cite pages like the second one and never the first, the gap is worth closing.

What to check: for each service you want to be recommended for, is there one page that says, in its first paragraph, what you offer, for whom, where, and what makes it different? Would it answer the question a customer actually types?

2. The sources for your trade aren't your website

Which pages ChatGPT relies on depends on the trade, and you can't know it without looking.

For dental implants in Tulsa, the answers leaned on practices' own websites, plus healthcare directories and professional associations. When we tried the same method on a tire shop in Tulsa (30 October 2026), the questions about mobile tire repair were answered mostly from Google business profiles and small hosted sites, not from businesses' websites. The shop we looked at calls itself a mobile tire and roadside service and has a dedicated page for mobile tire repair. It was named in 5 of 5 answers for 24-hour roadside help, but in only 1 of 5 for mobile tire repair, and its mobile repair page was cited in none of the 15 answers.

What to check: ask your three most important questions and look at the sources under each answer. If they are mostly business profiles, your profile's categories, services and description matter more than another page on your site. If they are directories, check whether you are listed there, under the right service, with the right address.

3. Someone else has more specific evidence for that exact question

"Being recommended by ChatGPT" is not one thing. It is a separate result for every question.

The practice in pattern 1 that did well on full-arch questions was named in 0 of 5 answers to "who is the best implant dentist in Tulsa", "I lost a front tooth, where should I get an implant" and "board-certified implant dentist in Tulsa". On those questions the answers cited a professional association's directory, a credentialing board, a healthcare directory and review aggregators. The evidence that wins "best" and "board-certified" is different from the evidence that wins "same-day implants".

What to check: don't test one question. Test the five or eight your customers really ask, including the plain ones ("best … in [city]") and the specific ones (a treatment, an emergency, a price).

4. Its search crawler can't read your site

ChatGPT's web search uses OpenAI's own crawler, OAI-SearchBot. OpenAI's documentation for site owners explains that a site can allow it to appear in search results, separately from the crawler used for training. If your robots.txt blocks it, or your security service blocks unknown bots, your pages can't be read for answers.

Two other things make pages hard to use: important facts that only appear inside images, and pages that show nothing until scripts run.

What to check: open yoursite.com/robots.txt and look for OAI-SearchBot. Ask whoever runs your website whether a firewall or bot filter blocks crawlers. Make sure your services, address and opening hours are written as text on the page.

5. You weren't in that sample

The same question can get a different answer a minute later. Results also depend on the model: in our Tulsa test, a healthcare directory was cited in 58% of a smaller model's answers and in about a fifth of gpt-5.4's, and the leader changed between the two.

So a single check in your own ChatGPT tells you little, in either direction. Being named once doesn't mean you are safe; being missed once doesn't mean you are invisible.

How to check your own business

You can do a rough version yourself in ten minutes:

  1. Open ChatGPT in a temporary chat, so your own history doesn't colour the answers.
  2. Ask five questions your customers ask, each with your city in it: one "best … in [city]", one "who do you recommend for …", and three about specific services.
  3. Ask each question three times. Write down whether you were named, who was named instead, and which sources were linked.
  4. Open the sources. Note what they say about the businesses that were named, and what they say about you, if anything.

Or let us do it: the free snapshot asks ChatGPT eight such questions for your business through OpenAI's API, with web search on and the location set to your city, and shows who it named and which pages it relied on, in about a minute. How we measure explains the setup and its limits.

What to fix first

In order of what we'd check first, not of what is guaranteed to work:

  1. One clear page per important service. First paragraph: what, for whom, where. Then the details a customer would ask about, in their words.
  2. The sources your answers already use. If the answers cite business profiles, complete yours: categories, services, hours, description. If they cite directories, claim and correct your listings.
  3. Reviews where the answers look. Ask happy customers to leave one, in the places your answers cite. Never pay or offer anything for them: Google's rules on fake engagement forbid incentivised reviews, and they can be removed.
  4. Let the crawler in. Allow OAI-SearchBot in robots.txt, and keep key facts in text.
  5. Measure again, on the same questions. Pages take time to be re-read. Compare like with like: the same questions, the same city, enough answers to tell a change from noise.

What not to do

Questions

How long does it take to show up after a change? There is no fixed time. Pages need to be re-crawled and then picked for an answer. We plan in weeks for a first measurement and in months for a stable change, and we say plainly if nothing moved.

Is this the same as SEO? It overlaps: clear pages, correct listings and reviews help both. The difference is that an AI answer names a few businesses and links its sources, so you can see exactly which pages it relied on for each question.

Does it work the same in every city and trade? No. The sources differ by trade, as the dental and tire examples show, and by place. That is why we measure first and decide what to fix from what the answers actually cited.

ChatGPT is a trademark of OpenAI; Talsara is not affiliated with OpenAI. The figures in this article come from our own stored answers; How we measure describes the setup.

More notes