How common are answers naming ten or more businesses?

Across the 926,792 answer units classified in this study, drawn from 193 query groups over the window 2026-04-27 to 2026-09-18, 39.4% of all answers named ten or more distinct businesses. That makes names 10 or more the second largest of the four classes measured, behind names 6 to 9 at 51.8% and ahead of names 1 to 5 at 6.6% and names none at 2.3%.

The pooled figure hides a real split between engines. ChatGPT put 49.7% of its 329,786 answers into this top class. Claude put 43.4% of its 298,968 answers there. Gemini put 24% of its 298,038 answers there, the lowest of the three engines measured. So a single pooled percentage for names 10 or more understates how differently the three engines behave: ChatGPT is roughly twice as likely as Gemini to produce a long list, with Claude in between but closer to ChatGPT than to Gemini.

There is also meaningful variation across query groups, separate from the engine split. Looking at how the names 10 or more class is distributed across the 188 query groups counted, the share of answers in this class ranges from a minimum of 5.7% in the least list-heavy group to a maximum of 73.6% in the most list-heavy group, with a median of 38.3%. That is a wide spread: some query groups almost never produce a long list from any engine, and others produce one more often than not. This suggests that the query itself, not just the engine, is a strong driver of list length. A query that invites an exhaustive answer, such as one asking for a broad category of options, will push any engine toward the names 10 or more class; a narrower query will not, regardless of which engine answers it.

What should a reader take from a single number like 39.4% of all answers? Mainly that it is an average across a genuinely heterogeneous set of engines and query types, and that averages here conceal more than they reveal. A business trying to understand how often it might be competing against nine or more other named businesses in an AI answer needs to know which engine and which kind of query it cares about, not just the pooled rate. The pooled rate is a reasonable starting point for describing the dataset as a whole, but treating it as the answer to a specific question about a specific engine or query type would be a mistake the breakdown above is meant to prevent.