Claude

Claude is the family of large language models built by Anthropic and the assistant of the same name at claude.ai. With web retrieval on, it fetches pages to answer a question and can name and link businesses in the result, which is what makes it one of the engines an AI visibility measurement covers.

What the name covers

One word does two jobs. Claude is the model, offered through an API in several sizes with different speed and cost, and Claude is the consumer assistant built on it. What both share is a context window: everything the model can reference while writing a reply, including the reply itself. Anthropic’s context window documentation puts the current models at either 200,000 or one million tokens, and notes that a retrieved page, a tool result and the conversation so far all count against the same limit. A page that reaches the model as text is competing for that space with everything else in the conversation.

Claude in this site’s measurements

Claude is the one assistant this site has probed at scale. For 9,240 small businesses, one buyer question from their market and trade was put to Claude (Opus) with retrieval on, and the answer was checked for the business’s name. 14.9% were named; among the 5,928 answers that gave a differentiated list rather than a hedge, 17.7%. The probe is written up in what 9,483 small business websites serve to an AI crawler.

Two limits attach to that number. It is one question per business on one day, a tenth of the panel used for paying clients, so it is directional. And it is a Claude number: it says nothing about ChatGPT, Gemini or Perplexity, which fetch, weigh and cite differently, and it should not be quoted as an AI rate in general.

A common question

Does the 14.9% figure apply to ChatGPT, Gemini and Perplexity as well?

No. That reading came from putting one buyer question to Claude with retrieval on, so it describes Claude and nothing else; the other assistants fetch, weigh and cite by their own rules, and the figure should not be repeated as a general AI rate. It is also one question per business asked on a single day, a tenth of the panel a paying client gets, which makes it directional rather than a settled measurement of any business's standing.

Sources

Context window behavior and sizes: Anthropic’s context window documentation. Probe figures are this site’s own, computed from its audit records; one assistant, one question per business, sample drawn per metro and trade.