Why ChatGPT recommends some local businesses and not others
Research · By Aashir Abdullah, founder of Taelvix · 20 November 2026
On 18 November 2026 we asked ChatGPT for the best dentist in the towns of 43 dental practices that US marketing agencies list as their clients. ChatGPT named only 9 of the 43, and only 2 of them first. When we looked closely at one practice it skipped, the causes were ordinary and fixable: contradictory facts on its own website, missing listings on the sites ChatGPT cited, and fewer reviews than the practices it recommended. The same checks apply to any local business.
How we ran the test
We took 43 dental practices named as clients on the websites of US marketing agencies. For each one we asked ChatGPT a single question a customer might ask, with the practice's town and state included, in a fresh session. We matched the question to the practice's specialty:
| Question asked | Practices |
|---|---|
| What is the best family dentist in [town]? | 28 |
| Who is the best pediatric dentist in [town]? | 4 |
| Who is the best orthodontist in [town]? | 4 |
| Who is the best cosmetic dentist in [town]? | 4 |
| What is the best dentist in [town]? | 2 |
| Who is the best oral surgeon in [town]? | 1 |
We recorded whether the practice was named, its position in the answer, and which businesses ChatGPT recommended instead.
What ChatGPT said
| Result | Practices (of 43) |
|---|---|
| Not named at all | 34 |
| Named first | 2 |
| Named second | 4 |
| Named third | 2 |
| Named, but listed after specialists | 1 |
In other words, 79% of these practices were left out of the answer to the question that matters most to them, and every one of them is listed as a client of a marketing agency.
One question asked once is a snapshot, not a verdict. AI answers change between runs. But a business that is absent from the most obvious question in its own town has a problem worth checking properly.
What we found when we looked closer
A one-question test shows who is missing, not why. To find out why, we ran a full audit of one practice that ChatGPT, Gemini and Google AI Overviews all left out. It has 4.9 stars from more than 400 reviews and a fast, crawlable website. Four things stood out.
1. Its own website contradicted itself
16 of its 48 pages showed an old opening schedule. Insurance, review counts and awards were described differently on different pages. When the facts conflict, an AI tool has less reason to be confident about the business, and Google's AI Overview repeated the wrong hours to customers.
2. It was missing from the sites AI cited
The answers that recommended other businesses cited ranking sites, review platforms and map listings. The businesses recommended most often appeared on those sites. The audited practice did not.
3. Competitors had more reviews
Its 400+ reviews were well below the 1,000 to 1,600 that most of the recommended practices had.
4. Nobody was named behind its health content
Its articles were credited to a website username instead of the dentist, which gives AI tools less reason to treat them as expert content.
That is one practice, so treat these as likely causes rather than rules. They are, however, the same signals AI tools can read in any industry: consistent facts, third-party mentions, reviews and named expertise.
What agencies can check this week
- Ask ChatGPT, Gemini and Google the obvious customer question for each client, with the town included, and write down who is named.
- Search the client's own site for its opening hours, service wording and review count, and make every page say the same thing.
- Look at the sources the AI answers cite, and check whether the client is listed there with the same name, address, phone number and hours.
- Compare the client's review count with the businesses that were named.
- Put the expert's name and credentials on every health or advice article.
Our GEO audit method covers these checks in full, and our white-label GEO services do them under your agency's brand.
Questions agencies ask
How can I check whether ChatGPT recommends my client?
Open a fresh ChatGPT session and ask the question a customer would ask, with the town and state included, for example "What is the best family dentist in Wheaton, Illinois?". Note whether the business is named and in what position, and which businesses are named instead. Then ask Gemini and Google the same question, because the answers often differ.
Does one test prove a business is invisible in ChatGPT?
No. A single question asked once is a snapshot. Answers change between runs and over time. It is a useful warning sign, and a full audit repeats 20 to 30 questions across three platforms before drawing conclusions.
How long do fixes take to change what ChatGPT says?
We don't have enough client history to give a reliable timeline yet, so we don't promise one. Some fixes, such as correcting opening hours on the website, take an hour to make. How quickly AI answers reflect them varies, which is why we re-test every month.