In short. Grok web search retrieves web pages, while xAI records encountered source URLs separately from inline links in the answer.

Definition: As of October 2026, Grok web search is xAI's documented capability for searching the web and reading pages through the web_search API tool. It retrieves information that Grok can use in answers with source links.

Key facts

xAI's Web Search guide describes real-time searching, page access and extraction of relevant information.
The Citations guide separates the complete list of sources encountered from links inserted into the answer text.
The separate X Search tool searches posts and users on X and fetches conversation threads.

How does Grok retrieve web information?

A tool is an operation the model can call while processing a request. xAI's Tools Overview describes a loop: the model analyses the request, chooses a tool call or a final answer, processes results, and continues until it has sufficient information. Built-in tools execute on the service's servers.

The Web Search guide provides this tool configuration within a request:

{
  "tools": [
    { "type": "web_search" }
  ]
}

The overview's example includes both web_search and x_search in the same request. These are separately configured tools for web pages and X content. The documented workflow can therefore combine web retrieval with social content when both tools are supplied.

Which controls affect the pages searched?

The Web Search parameters include allowed_domains, which restricts searching and browsing to specified domains, and excluded_domains, which removes specified domains from web search calls. Each list accepts up to five domains, and the same request cannot use both settings together.

These settings belong to the API request. For a website review, the configured domain list is therefore relevant evidence about which pages the tool can consider. A test restricted to a particular domain measures retrieval within that restriction.

The same guide documents enable_image_understanding, which lets the model inspect images encountered during browsing. Page retrieval can consequently involve visual content as well as text when that setting is enabled.

How are retrieved sources and answer citations recorded?

In xAI's Citations documentation, response.citations lists URLs encountered during the search process. A source can remain in this list even when the model judges it insufficiently relevant to reference in the final answer.

Inline citations are source links placed in the answer text. When enabled, structured citation data in an annotations array includes the URL, character positions and citation label. This connects a displayed link to its position in the response.

For an audit of a website's appearance in an answer, this distinction supports separate records for retrieval and visible attribution. An encountered URL and a link attached to the written answer describe different stages of the documented process.

FAQ

Can you treat every returned URL as an answer citation?

The Citations guide says the complete list includes sources that the final answer may leave unreferenced. Check the answer's inline links when recording visible citations.

Can you restrict a search to particular websites?

The Web Search guide documents allowed_domains for restricting searching and browsing. This is an API configuration available to the application making the request.

Are inline citations enabled automatically?

The Citations guide documents different defaults: the Responses API enables inline citations by default, while the xAI Python SDK requires include=["inline_citations"]. The model decides where and when to insert links according to the request.

Sources

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