Whom AI assistants name when asked where to book an appointment

"Where can I book" is a question people used to ask a search engine and now ask an AI assistant more and more often. We decided to see whom it names in reply, and we asked ourselves: 33 days in a row, 103 questions about booking a doctor or a specialist, 1,965 trials, from July 27 to September 3, 2026. We analyzed each answer and recorded whose website it cited

Whom do AI assistants name most often?

Maps hold first place in our measurement by a wide margin. Second place goes to a medical directory. Next come websites built to be read by a machine: their pages hold opening hours, services, and a way to book in a form the assistant parses without a human

One number in the table can't be read at face value

The three top rows were collected in different ways, and this matters more than the order itself

The assistant names 2GIS and the directory on general questions: manicure in Almaty, dentistry nearby. It names our address on named ones, when a person asks about a specific specialist or clinic by name. Of 103 questions, we were named in 35, and in each of them the assistant opened the page of a specific business. It never named the catalog shelf "city plus service"

So a place in the table does not say that we are preferred. It says that we have many pages of specific businesses, while general questions have their own owners

What this means for a business

A question a person asks an assistant does not sound like a search query. Instead of "dentistry Almaty" they write "where to have a child's teeth checked on Saturday in Almaty, how much does it cost." The assistant answers in connected text and names two or three sources

Getting into that answer and getting into search results are different jobs. In search results, links and the age of the site decide. In an assistant's answer, what decides is whether the page can be parsed by a machine: whether it has hours by day, service names in words, the city, and a way to book without a call

How we measured

Each question was asked daily, the answer was parsed into a list of sources, and the sources were added up over the whole period. We counted mentions of a domain, and clicks are not counted: the assistant doesn't pass them on, it names the source in the text of the answer

The numbers are ours, and the method is simple and repeatable. If you need the full table for all 103 questions, write to us and we will send it