Do ChatGPT, Gemini and Claude recommend the same businesses? We asked all three 10 times

How did we compare ChatGPT, Gemini and Claude?

We asked each assistant three questions a customer might type, word for word, 10 times each: "Who is the best dentist in London?", "Can you recommend a good plumber in Chicago?" and "What is the best hair salon in New York?". That is 90 answers, collected on 13 October 2026 through each maker's API with web search available: ChatGPT (gpt-5.6-terra), Gemini (3.1 Flash-Lite) and Claude (Sonnet 5). Each assistant decided for itself whether to search.

We counted every business named, matching spellings so that one business counts once, and left out directories and marketplaces such as Angi and Nextdoor. No location or chat history was involved, so the differences come from the assistants alone. The apps can add both, which only widens the gap between two customers; read this as how the three models behave, not as a screenshot of what one person sees (how we test).

It is a small, focused test: three questions, three cities, one day, and the lighter Gemini model Peekl uses in its audits. It shows how far the assistants can diverge on an identical question, not a ranking of who is right.

Who did each assistant name for the same question?

Mostly different businesses, even when all three answered every time. The table lists, for each question, how many of the 10 answers named at least one business and which businesses came back in at least half of them. These regulars are the closest thing each assistant has to a favourite, and they barely overlap.

For the London dentist, no assistant's regular was a regular for another. Harley Street Dental Clinic, named in 7 of Gemini's answers, appeared once on ChatGPT and once on Claude. For the New York salon, Suite Caroline came up in 9 Gemini answers and 7 Claude answers but in only 1 of ChatGPT's. The plumber is the only question where ChatGPT and Gemini shared regulars: both named Vanguard Plumbing & Sewer and Rescue Plumbing in at least half of their answers.

How many businesses did all three assistants name?

Very few: 8 of the 122. Three of the 30 London dentists were named at least once by every assistant, none of the 25 Chicago plumbers and 5 of the 67 New York salons. Being named by one assistant only was the norm, for 86 businesses, and even those eight were rarely steady: only Suite Caroline came back in at least half of the answers on two assistants.

The salon question produced the most names, 67, because Claude listed up to 14 salons per answer. More names did not mean more agreement: 44 of those salons were still named by a single assistant.

Why did Claude name no plumber?

In this test, every answer in which Claude named nobody was one where it had not searched the web. Claude decides for each question whether to look things up, and it searched in only 2 of its 10 dentist answers, none of its plumber answers and 6 of its salon answers. In the 18 dentist and plumber answers without a search, it named no business at all; for the plumber it pointed to Angi, Nextdoor or the Better Business Bureau in 6 answers instead.

For the salon it named businesses even without searching, in 4 answers, but those were well-known, long-running names such as Sally Hershberger, Oscar Blandi and Frédéric Fekkai. Salons such as Fox & Jane, Miss Rizos and Bloom Beauty Lounge came up only in answers where Claude had searched. So for most local businesses, being named by Claude depends on two things: Claude choosing to search, and then finding you on the pages it reads. How to check if Claude recommends your business covers what those pages are.

In the Claude app, people can also turn web search on or off, which changes the same pattern again. That is one more reason to test Claude separately rather than assume it follows ChatGPT.

Do ChatGPT and Gemini agree with each other more often?

Not much more. Across the three questions in this test, ChatGPT and Gemini had 17 businesses in common, against 21 for Gemini and Claude and 14 for ChatGPT and Claude. The larger picture comes from our earlier study of 756 answers about seven trades in nine cities: for the same trade and city, ChatGPT and Gemini named at least one business in common only 51% of the time.

They agreed far more on what to read than on whom to name: in 76% of those cases they cited at least one website in common. The same directories and review sites fed both, and each still reached its own shortlist. Which websites AI trusts lists those sources and explains the method.

What does this mean when you check your own business?

Check each assistant on its own, and more than once. Someone who asked ChatGPT once for a London dentist would most likely have seen Serio Dental, named in 6 of its 10 answers, with no hint that Gemini's regulars were three other practices or that Claude usually named nobody. One answer from one assistant is a sample of one. Our larger Gemini study found the same on one assistant alone: asked three times, only 1 in 4 named businesses came back every time.

Then read the gaps per assistant rather than as one score: on ChatGPT and Gemini it is about which pages each one reads, on Claude also about whether it searches at all. The guides on ChatGPT, Gemini and Claude show how to check each by hand. Peekl's free check asks ChatGPT or Gemini twelve of your customers' questions; the full audit asks all three, three times each.