Why Does ChatGPT Recommend Different Businesses Every Time?

Why does ChatGPT recommend different businesses every time?

ChatGPT writes every answer from scratch. It does not keep a ranked list of the best dentists in your city. It generates a new response each time, usually after pulling in web sources, so the businesses it names shift from one ask to the next. In our data, 23% of the US dentists we scanned were named in at least one ChatGPT answer, but on average only 11% of the answers about a business actually named it.

That gap is the whole story. Being named once is not the same as being named reliably, and the difference changes how you should check your own visibility.

AI answers are generated fresh for every question, not looked up from a fixed list

A classic search results page is built from an index and a ranking. A large language model works differently. It produces text piece by piece, choosing each next word from a range of likely options, and a small amount of randomness is part of the design. Two runs of the same prompt can come back with different phrasing, a different order and different names.

For a question with one correct answer, that variation rarely matters. For "best dentist in Austin", where dozens of practices are reasonable picks, the model can land on a different handful each time.

A single answer is one sample from a range of possible answers.

Why do live web sources make the answer shift?

The second cause is retrieval. ChatGPT leaned heavily on sources when answering questions about dentists: 100% of the ChatGPT answers we analysed for dentists cited at least one link, and they averaged 32.6 cited links per answer.

When the engine searches before answering, the pages it pulls in can differ by wording, by location and by day. "Best dentist in Austin", "top-rated family dentist Austin" and "emergency dentist open Saturday" may retrieve different review pages, directories and articles. Each set may favour different practices, and so each answer may name different ones. We did not test wording directly, so treat this as a likely mechanism.

Then there are changes nobody outside the engine can see: model updates, changes in how search results are fetched, and new reviews or articles appearing on the web. Any of these can move a business in or out of answers over a few weeks. Our snapshot does not isolate how much each factor contributes, so we won't pretend to rank them.

How consistently does ChatGPT name the same business?

Not very. We ask each question several times per business and record every answer separately. That gives us two different measures, and they tell very different stories. The sample is small: 31 dentists on ChatGPT and 29 on Google AI Mode, so read the figures below as directional.

Share of businesses we scanned that were named at least once

On ChatGPT, 23% of the dentists we scanned were named in at least one analysed answer. On Google AI Mode, 14% were. This is the generous measure. A single appearance in any answer counts.

Average share of a business's answers that named it (mention rate)

Mention rate is the share of a business's AI answers that name it. We calculate it for each business, then average across businesses. On ChatGPT the average was 11%. On Google AI Mode it was 7%.

If every dentist who appeared once had appeared in every answer, the two figures for each engine would match. They are far apart on both engines. So even the practices that did get named were missing from many other answers to the same kinds of questions.

In practice, a dentist who asks ChatGPT once and sees their name could ask again the next day and not see it. Both results would be accurate. Neither one alone tells you much.

What the data does not show: which individual practices were consistent and which were one-off appearances, or how mention rates are spread across businesses. We report group averages only. For the full dentist breakdown, see how often ChatGPT recommends dentists.

Do ChatGPT and Google AI Mode recommend the same businesses?

Often they don't. The variation is not limited to repeated asks on one engine; the two engines we compared also disagree with each other.

Combined reach vs. each engine alone

Across both engines combined, 29% of the dentists we scanned were named at least once. That is higher than ChatGPT alone at 23% and Google AI Mode alone at 14%.

If the two engines picked the same practices, the combined figure could be no higher than the larger of the two. It is higher, which means some dentists showed up on one engine and never on the other. The groups differ slightly in size (31 businesses checked on ChatGPT, 29 on Google AI Mode), so read this as evidence that the engines diverge, not a precise overlap measurement. The sample is small (31 dentists), and the gap rests on a few practices.

Consistency stays low when you pool the engines, too. The average mention rate across both was 9%.

How many sources each engine cites per answer

The engines also build their answers from very different amounts of material. ChatGPT answers for dentists averaged 32.6 cited links. Google AI Mode answers averaged 14.0, and 96% of them cited at least one source, against 100% for ChatGPT.

Citing different amounts of material, and possibly different sites, is a plausible reason the two engines name different practices. Our data shows the difference in citation volume; it does not prove that this difference causes the difference in who gets named. We compare the two engines in more depth in ChatGPT vs Google AI Mode for local business recommendations, and we break down which sites get cited in what sources ChatGPT uses to recommend dentists.

What this means when you check your own business

Why one test question proves nothing

For the average dentist in our data, any single ChatGPT answer was far more likely to leave them out than to name them. So a single "no" is a common result even for a practice that does get recommended some of the time, and a single "yes" is no proof of reliable visibility either.

It is also a likely explanation when ChatGPT seems to stop recommending your business: it may not have stopped. You saw one answer that included you and a later one that didn't, and both could be draws from the same shifting pool. Before concluding anything changed, ask the same questions several times, in a few different wordings, over more than one week.

Testing from your own logged-in account adds noise of its own, because chat history and location can shape the answer. Our step-by-step guide on how to check if ChatGPT recommends your business covers how to run a cleaner manual check.

Mention rate vs. 'did it show up once'

The two measures answer different questions. "Named at least once" tells you whether the engine surfaced you at all for those questions. Mention rate tells you how often a customer asking that question would actually hear your name.

We think mention rate is the number that matters for a business, because a customer asks once. They don't rerun the question until your name appears. A practice named in a small share of answers is reachable in theory and mostly invisible in practice, and only a rate measured repeatedly over time can show whether that share is growing or shrinking.

How to become a business AI names more consistently

A caveat first: our snapshot does not measure which actions raise a business's mention rate, and no one can guarantee a place in AI answers. What we can say is that both engines cited sources in nearly every dentist answer we analysed, so the practical lever is being present, accurate and well described on the sources they read.

1. Find out which sources get cited for your questions. Look at the links under the answers for your city and service, and note which directories, review sites and articles keep coming up.
2. Complete and correct your Google Business Profile. Categories, services, hours, photos and description should be accurate and specific. Google AI Mode is Google's own product, so it is reasonable to expect your Google Business Profile to matter there, though our data does not measure this.
3. Keep your details consistent everywhere. Name, address, phone, services and insurance information should match across your site, Google Business Profile, Yelp and the directories your patients use.
4. Write pages that answer the questions customers ask. A clear page per core service, with your city, what's included and who it's for, gives an engine something concrete to cite.
5. Earn reviews that describe specific services. Reviews that mention implants, emergency visits or children's dentistry give engines more to match against specific questions.

How we got this data

The figures come from US dental practices in ClientRadar's data, combining businesses tracked with weekly runs and businesses checked with the free scan. ChatGPT figures cover 31 businesses and Google AI Mode figures cover 29 businesses, from July 20 to October 11, 2026. Citation figures cover answers given for 32 businesses on ChatGPT and 30 on Google AI Mode over July 20 to October 11, 2026; an answer reused for two businesses counts once.

We asked each engine the questions customers ask, several times per business, through the engines' official APIs, and recorded which businesses each answer named and which links it cited. API results closely approximate but can differ from what the consumer apps show, for example because the apps use chat history and location. These numbers describe the dentists we scanned, not all dental practices, and they cover two engines only.

The practical takeaway

Stop judging your visibility by a single question. This week, ask your three most important customer questions several times on each engine, write down the links they cite, and start a weekly log of how often you are named. After a few weeks, that log will tell you far more than any one answer. If you want a starting point, you can run a free scan of your business, which asks ChatGPT and Google AI Mode three customer questions about you.