In short. Agentic search uses results and retrieved documents to guide further searching and reading within a task.

Definition: Agentic search is an iterative process in which an AI assistant uses search results and retrieved documents to decide what to search for or read next while completing a task. The assistant can make successive searches and examine selected pages before producing an answer.

Key facts

As of 17 October 2026, Anthropic's web search tool documentation describes searches that can repeat within a single request to the Claude API, Anthropic's programming interface.
Anthropic's web fetch tool documentation describes retrieving the full content of web pages and PDF documents, including pages located through search.
In Building effective agents, published on 19 December 2024, Anthropic describes agents as systems that direct their own processes and tool use.

How does the assistant choose its next step?

A tool is a capability an assistant can invoke, such as searching the web or retrieving a page. Anthropic's Building effective agents describes a loop in which tool results provide feedback for the next action. Its discussion of search includes repeated searching and analysis, with an evaluator deciding whether further searches are warranted.

Anthropic's combined search and fetch example makes the sequence concrete: Claude finds relevant articles, selects results, retrieves their full content and produces an analysis. Search supplies candidate sources; fetching supplies the document content for examination. A result can therefore lead to a further reading step within the same task.

How is the search loop controlled?

Anthropic's web search tool definition includes this configuration, shown here without its optional domain and location settings:

{
  "type": "web_search_20250305",
  "name": "web_search",
  "max_uses": 5
}

The version identifier selects a documented search tool. The max_uses setting caps searches per request at five in this example. If the assistant attempts another search beyond the cap, the tool returns a max_uses_exceeded error. The number of searches performed depends on the task and the assistant's decisions within that limit.

What matters for a website?

The retrieval tool affects which page content the assistant can examine. Anthropic's web fetch documentation states that its fetch tool does not support sites rendered dynamically with JavaScript.

For a website owner, this implies checking whether the relevant text is available through the retrieval method used by the assistant. The same documentation distinguishes fetching from browser use, where an application operates a browser and returns page text or screenshots. Access therefore depends on the configured tools as well as the page itself.

FAQ

Can you limit the number of searches?

Yes. Anthropic's web search documentation defines max_uses as a limit on searches per request. The configuration above permits up to five searches.

Does browsing always involve a browser?

In this context, browsing includes retrieving and reading page content. Anthropic's web fetch tool retrieves documents directly; its documentation describes browser use separately for interactions such as clicking or filling forms.

What stops an agentic search task?

Anthropic's agent design guidance describes completion and explicit stopping conditions, including a maximum number of iterations. A search tool's usage cap provides an additional limit on search calls.

Sources

Check how a page reads to answer engines. Open the GEO score analyzer