xAI · Answers
Grok-DeepSearch
One prompt, tens of fetches: that is the shape a deep-research run leaves in a log, and the mode itself is documented rather than invented. xAI's multi-agent docs describe orchestrating several agents that search and gather information from multiple sources simultaneously and return comprehensive, well-sourced answers with citations and supporting evidence. What is undocumented is the request signature. xAI nowhere states that those fetches carry this token, or carry any name at all, so the pattern is firmly established and the label attached to it is not.
- Operated by
- xAI
- Purpose
- Answers
- robots.txt token
- Grok-DeepSearch
- Verification
- User agent only
Fetches a page in real time to answer a question someone just asked.
How to verify Grok-DeepSearch
xAI publishes no list of IP addresses for Grok-DeepSearch, so nobody (us included) can prove that a request carrying this user agent really came from xAI. The name is trivial to copy. Treat it as a claim the visitor is making about itself, not as an identity anyone has checked.
There is no range list to confirm. This entry was last reviewed against xAI's own documentation on .
User agent
xAI has not published a full user-agent string for Grok-DeepSearch. Requests are identified by the Grok-DeepSearch product token appearing in the User-Agent header; we match that token rather than a whole string, because the rest of the header varies and matching it would miss real traffic.
robots.txt for Grok-DeepSearch
Block
User-agent: Grok-DeepSearch
Disallow: /Allow
User-agent: Grok-DeepSearch
Allow: /robots.txt is a request, not an enforcement mechanism. It is honoured by convention, and a crawler that ignores it is stopped at your edge, not in a text file.
What blocking Grok-DeepSearch costs you
Exclusion here trims the long tail rather than the headline. A multi-step research pass reaches reference pages, comparison tables and documentation that a single-shot answer lookup would skip, so what disappears is the secondary citation rather than the primary one. Cutting the other way is a load argument no other entry in this group can make: if these bursts are real, they are by some distance the most expensive traffic here per user query, which gives an operator with a tight crawl budget a reason to deny that has nothing to do with visibility. Whether a robots.txt rule reaches them is unknowable, because xAI has never stated a robots.txt policy for anything it runs.
Vendor documentation
xAI does not publish documentation for Grok-DeepSearch that we could find. Everything on this page comes from what they do publish elsewhere and from observed behaviour, so treat it accordingly.