How to Appear in Grok and Get Your Brand Mentioned

Grok is the assistant built by xAI. According to the registration statement SpaceX filed with the US Securities and Exchange Commission on 20 May 2026, Grok Chat is available on grok.com, in dedicated mobile apps, inside X and through the xAI API, with real-time integration of web search and X data.

The same filing gives the scale. Across Grok and X there were approximately 550 million monthly active users, and approximately 117 million of them had used Grok’s AI features as of 31 March 2026. For a company, that is an audience large enough to be worth checking, reached through an assistant that lives inside a social network.

The practical consequence of the definition is that there is no switch to flip. Grok decides which brands to name when it writes the answer, and a company influences that decision only through what is published about it, on its own site, on X and on other people’s pages.

The training data has a date. xAI gives May 2026 as the knowledge cut-off of Grok 4.7, the model in its current documentation, consulted on 22 October 2026. What the model knows about a company without searching comes from what was published before then, mostly on sites the company does not control.

Web Search, in xAI’s words, lets Grok search the web in real time and browse pages to extract information. X Search is a separate tool that runs keyword search, semantic search, user search and thread fetch on X. The two are called independently, so a company can be visible in one and absent from the other.

Search does not always produce a link. xAI’s documentation on citations says that turning them on does not mean the model will cite sources in every answer, and that not every URL it looked at is referenced in the final text. A brand can therefore shape an answer without being linked, which is why mentions matter on this engine.

Read across the rows, the table explains why a fix on the website moves only one route out of three. A company that rewrites its pages improves what the Web Search tool can extract, but the model’s memory changes only with the next training run, and X Search depends on what is posted on X. How each assistant decides which sources to cite is covered surface by surface in how AI models choose their sources.

The same filing gives the volume Grok can draw on: approximately 350 million posts a day across X. A company that never posts is absent from that stream, and whatever Grok finds about it on X is what other people wrote.

X Search works at the level of accounts. xAI’s documentation lets a developer restrict X Search to up to 20 named handles, or exclude up to 20, which shows that the tool reasons about who posted, not only about what was posted. One official handle, with the same company name and the same one-line description as the website, is easier for Grok to connect to the brand than several half-used accounts.

X is not, however, where most of Grok’s links point. In a measurement Ahrefs published on 1 June 2026, covering more than 1.9 million US queries, x.com ranked twelfth among the domains Grok cited, with 1.4% of the citations. The useful reading is that X shapes what Grok knows and says about a brand, while the links in its answers still go mostly to other websites.

The figures come with limits. They cover US queries in one month, they are measured by a company that sells AI visibility tracking, and the shares are calculated only over the 50 most-cited domains. They show where Grok’s links tend to go, not what it says about any one brand.

For a company, the list points away from its own website. Most of what Grok cites is written by users on large sites, which is why the work on this engine is as much about being mentioned on other people’s pages as about publishing.

Any name for a Grok robot found in third-party lists is therefore a guess, not a documented fact. A company should not build rules around a name the provider itself does not confirm.

What a company can still check is that nothing blocks robots in general. A bot-protection rule at the firewall or at the CDN, the layer that serves a site before the server does, can stop any automated visit without anyone deciding so. Checking whether AI robots can read the site shows whether the pages are reachable at all, which is the condition for the Web Search tool to read them.

Visits are the easier part. According to Google Analytics help, consulted on 22 October 2026, Google Analytics 4 groups visits from assistants such as ChatGPT, Gemini, DeepSeek or Grok into a channel called “AI Assistant”. That channel only counts people who clicked a link in an answer.

Server logs help less here than on other assistants. The standard method is a rule at the firewall or the CDN that logs, without blocking, the requests from each AI provider’s robot, identified by its user agent and, where the provider publishes its addresses, verified by IP. This is what DoGeo sets up for each client, with a monthly count per robot and per page. With no documented name, Grok’s visits cannot be isolated in those logs today.

What neither source shows is what Grok writes. Whether it names the company, how it describes it and which page it links are only visible by asking Grok the same questions every month, on grok.com and on X, and recording the answers. That is slower than a report, and it is the measurement this engine allows.