How does Claude web fetch differ from web search?

Definition: As of 17 October 2026, Claude web fetch retrieves content from a specified web page or PDF document, while web search finds sources through a search query. A workflow can use search to locate a document and fetch to read its full content.

Key facts

• Anthropic's web search tool documentation describes searches that can repeat within an API request, followed by an answer with cited sources.
• Anthropic's web fetch tool documentation restricts fetchable URLs to addresses already present in permitted conversation content, including earlier search results.
• Anthropic's Claude Help Center guide also describes web fetch in chat: when Web search is toggled on, Claude can read pages from supplied URLs.

How do search and fetch work together?

A search request starts with information to find. Anthropic's search documentation lists current prices and recent announcements among the reasons Claude searches. Claude decides when to search from the prompt, and the API runs the searches and returns results.

A fetch request points to a particular resource. The fetch documentation gives the example of a user asking Claude to read the README from the anthropics/anthropic-sdk-python repository without supplying its URL. With both tools enabled, Claude searches for the resource, then fetches it. For a website review, supplying the exact page URL therefore identifies the page to read; a broader search request leaves source selection to the search workflow.

How do citations differ in the API?

Citations are references connecting an answer to source material. Anthropic's fetch documentation states that search citations are always enabled, while fetch citations are optional and disabled by default. The following API request fragment enables both documented basic tool versions and turns on citations for fetched documents:

{
  "tools": [
    {
      "type": "web_search_20250305",
      "name": "web_search"
    },
    {
      "type": "web_fetch_20250910",
      "name": "web_fetch",
      "citations": {"enabled": true}
    }
  ]
}

The same documentation describes a fetch result containing the fetched URL, a document block and a retrieval timestamp named retrieved_at. These fields identify the retrieved resource and its retrieval time.

What should a website owner check?

Anthropic's fetch documentation states that the API tool does not support websites dynamically rendered with JavaScript. A page review therefore needs to establish whether its relevant text is available without that rendering step.

The documentation also says that fetch results can come from a cache, so returned content may differ from the latest page. For supported tool versions, "use_cache": false requests fresh content. This setting matters when checking a recently edited page. The max_content_tokens setting can truncate fetched text; its limit applies to text content and excludes binary content such as PDFs.

Related terms

• Claude web search
• Claude-User
• JavaScript rendering and AI crawlers
• AI citation

FAQ

Can I give Claude a URL directly?

Yes. Anthropic's chat help guide describes retrieving full page content from supplied URLs when Web search is toggled on. The guide warns that analyzing a long linked article can consume a significant portion of a free account's usage limit.

Can the API fetch a URL Claude invents in its answer?

Anthropic's URL validation documentation excludes URLs appearing only in Claude's own output or only in the system prompt. You can supply a URL in a user message; earlier web search results are also permitted sources of URLs.

Sources

1. Anthropic, "Web search tool", accessed 17 October 2026
2. Anthropic, "Web fetch tool", accessed 17 October 2026
3. Anthropic, "Enable and use web search", accessed 17 October 2026

---

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