What is Grok DeepSearch?
Grok DeepSearch is the xAI agent used for research questions that require more than a single web lookup. xAI describes DeepSearch as a system that seeks information across the web, compares conflicting facts and opinions, and produces a report for the person who asked the question.
Its defining behavior is multi-step retrieval. A research run can search for sources, open pages, extract relevant material, and synthesize the results. When Grok cites a page in the final report, that citation is tied to the answer being assembled rather than proof that the page entered a permanent search index.
A reachable page can be used as evidence in a DeepSearch response, so blocking the fetcher may reduce the chance that Grok quotes or links to that page during research. It does not guarantee that Grok will never mention the site, since the model can answer from other sources or existing model knowledge.
this bot classifies Grok DeepSearch as an AI Assistant, not as an AI Training crawler. xAI's public description covers live research and source synthesis. It does not say that requests carrying this token collect pages for model training, so this user agent should not be treated as a training-data control.
The published token in this directory is Grok-DeepSearch. Robots.txt support is partial, which means a disallow rule is useful as a preference but is not a dependable hard block for every request. Sites that must prevent access need an enforced rule at the server, application, or CDN.
Grok DeepSearch is marked unverifiable. The facts record provides no Cloudflare entry, signed-agent key directory, or verified IP range for this identity. A request can be classified by its token, but the header alone cannot prove that xAI sent it.
