AI names a local business 84% of the time when it has read the business's own page. 16% when it has not.

Key findings

• Businesses were named 5.2 times as often when the engine had opened one of their own pages before answering (83.7% vs 16.2%).
• The gap was largest on Claude: 86.6% vs 11.9%, about 7.3 times.
• Gemini named a business it had read 92.2% of the time, the highest rate of any engine.
• ChatGPT showed the smallest gap of the chat engines, 79.4% vs 21.1%, still about 3.8 times.
• Google AI Overviews named businesses it had not read more often than any other engine (33.4%), but still named a business it had read 2.7 times as often.
• On a first measurement, before any work, 70% of answers named a competitor or nobody (18,613 answers).

Why this matters for a local business

When someone asks an AI engine who to call, where to eat or which dealership to visit, the engine usually searches the web, opens a handful of pages, and writes its answer from what it read. A business whose own page is among the pages opened is in the room when the answer is written. A business that is not has to be carried in by someone else's page: a directory, a review site, or an article. Most of the time, it is not carried in.

The practical question for any local business is therefore not only whether its website ranks in a list of links. It is whether the AI engines open its pages when a customer asks the question it wants to be the answer to.

How the data was collected

• Questions are the kind real customers ask, about real local businesses in the United States, in the towns those businesses serve.
• Each question was put to ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews with live web search on, as a consumer would use them.
• Every answer was stored in full, together with the engine's own record of the web pages it opened or cited while answering.
• Each answer was then read for one thing: does it name the business being measured. Only answers read this way are counted.
• An answer counts as "read" when the engine's record of pages includes a page on the business's own website.

Limitations

• This is an observational comparison, not a controlled experiment. Part of the gap may run the other way: an engine may open a business's page because it already intended to recommend it.
• The engines only report some of what they read, so some answers counted as "not read" may have used the business's page without saying so.
• The businesses are ones Cloute measures, in the United States, across several local industries. Other industries and countries may differ.
• AI engines change their models and search behaviour often. These figures describe 2026.

Because the comparison is between answers to the same kinds of questions on the same engines, it is not affected by which questions were chosen or by a model update partway through the year.