Coaley Peak GEO Research Centre Engine guide
Grok How the assistant inside X searches posts and the web, and how a brand gets in.
Grok is the AI assistant from xAI, now SpaceXAI, built into X and available at grok.com and in apps. It searches public X posts as well as the web, and documents no crawler of its own. This guide covers its timeline, the four layers behind its answers, the crawler question, X's data settings and how to opt out.
Independent guide: Coaley Peak is not associated with SpaceXAI (formerly xAI). Trademarks
- Owner
- SpaceXAI (formerly xAI)
- Launched
- November 2023 (US early access through X Premium+)
- Home market
- Every country where X is available; grok.com and apps
- Languages
- Not published
- Open weights
- Grok-1 (Apache 2.0) and Grok 2 (community licence); not Grok 4.x
- Crawlers
- None documented by SpaceXAI
Overview
What Grok is, and who uses it
Elon Musk announced xAI in July 2023, and Grok launched that November, in early access for some US subscribers to X Premium+, with 'real-time knowledge of the world via the X platform'. xAI released the weights of the Grok-1 base model under Apache 2.0 in March 2024, launched Grok 3 with reasoning modes and DeepSearch in February 2025, and Grok 4 in July 2025. Grok 4.7, on a new, larger base model, arrived in September 2026.
Ownership has moved quickly. xAI acquired X in March 2025, SpaceX acquired xAI in February 2026, and xAI was rebranded SpaceXAI in July 2026; a further rename to SpaceXSI has been announced but not implemented. X Corp., which runs X, is a separate company in the same group.
A SpaceX filing with the US Securities and Exchange Commission in June 2026 put Grok at about 117 million monthly users of its AI features in March 2026, against about 550 million monthly users of Grok and X combined. Grok is available in every country where X is, and at grok.com and in apps with free and paid SuperGrok tiers.
For brands, Grok has one unique trait: public X posts are a first-party source for its answers, alongside the web. What a brand and its customers post publicly on X can feed Grok's answers directly. Regulators have taken note: Ireland's Data Protection Commission opened an inquiry into training on EU users' posts in April 2025, and Ofcom opened an investigation into Grok-generated imagery in January 2026.
Timeline
Grok, from launch to today
Elon Musk announces the AI company xAI.
xAI announces Grok, with real-time knowledge from X, in early access for a limited number of US users.
xAI releases the Grok-1 base model's weights and architecture under the Apache 2.0 licence.
xAI releases Grok-2 and Grok-2 mini in beta.
Grok 3 launches with reasoning ('Think') modes and DeepSearch.
xAI acquires X in an all-stock deal valuing X at US$33 billion.
Ireland's Data Protection Commission opens an inquiry into X's use of EU users' public posts to train Grok.
Grok 4 launches, with native tool use and real-time search.
xAI publishes Grok 2's weights on Hugging Face under the Grok 2 Community License.
Ofcom opens an Online Safety Act investigation into X over Grok-generated sexualised imagery; Indonesia and Malaysia had temporarily blocked Grok days earlier.
SpaceX acquires xAI, making it a wholly owned subsidiary.
xAI is rebranded as SpaceXAI.
SpaceXAI announces Grok 4.5, trained alongside Cursor, for coding, agentic tasks and knowledge work.
Grok 4.7 launches on a new, larger base model.
Things to know
Things to know about Grok
Public X posts are a first-party source for Grok's answers, alongside the web, so a brand's public activity on X can feed Grok directly.
SpaceXAI documents no crawler user agent or IP ranges for Grok. The 'GrokBot' string in some logs comes from a third-party directory, which classes it as undocumented.
Grok's web results come from 'internet search providers' that SpaceXAI does not name.
X's training opt-out does not stop Grok searching and surfacing public posts; only protecting an account does both.
Grok on X is governed by X's privacy policy and terms, and grok.com and the Grok apps by SpaceXAI's, so opt-outs and data requests differ by surface.
Grok's API returns every URL it encountered as a citation, including sources not used in the final answer, so appearing in its citations does not mean a page shaped the answer.
Grok conversations that users share get public URLs, which may be indexed by search engines such as Google.
xAI publishes Grok's system prompts on GitHub, so you can read how the assistant is told to search and cite (last updated November 2025).
API users can restrict Grok's web search to, or exclude, up to five domains, and its X search to or from up to 20 accounts.
Grok-1 and Grok 2 weights are public, but the current Grok 4.x models are not, and no knowledge cut-off is published for them.
The four layers
How Grok builds an answer
Every AI answer is made in four layers: what the model learned in pre-training, how it was tuned, what it retrieves live, and how it puts the answer together. Here is each one for Grok, and what it means for a brand that wants to be cited.
Layer 1
Pre-training: the base (foundation) model
Grok 4.7, released in September 2026, uses a new, larger base model than Grok 4.6; its weights are closed and no knowledge cut-off is published. Grok-1's 314-billion-parameter base model and Grok 2 are the only public weights. SpaceXAI says training data is primarily a large corpus of publicly available information, including raw web pages, metadata and text extracts, with quality filtering. X may share users' public posts and their Grok interactions for training unless they opt out, and grok.com conversations are used unless 'Improve the model' is switched off.
What it means for you: Grok learns from public X content as well as the web, so how your brand is discussed publicly on X can shape what Grok knows about you. Your own posts, and how customers talk about you, both count.
Layer 2
Post-training: tuning and alignment
SpaceXAI describes large-scale reinforcement learning: Grok 4 ran it at pre-training scale on the 200,000-GPU Colossus cluster, Grok 4.5 across hundreds of thousands of mostly technical tasks, and Grok 4.7 on a longer, harder run. X says Grok's pre-training data was reviewed and curated by human 'AI Tutors'. xAI publishes Grok's system prompts on GitHub.
What it means for you: The published system prompts show how Grok is told to search and cite. They are among the most direct insights any provider offers into how its assistant chooses sources, and we use them to shape content.
Layer 3
Retrieval: live search
Grok decides whether to search public X posts and run a real-time web search. SpaceXAI's privacy policy says public X posts are provided by X, and web data by internet search providers it does not name. Through the API, web search can be limited to, or exclude, up to five domains, and X search can filter by account, date and thread.
What it means for you: Two indexes matter: the web, through unnamed search providers, and X itself. Strong organic visibility covers the first; an active, credible presence on X, with posts that state your facts, covers the second.
Layer 4
The answer: what people read
Through the API, responses return a list of every URL the agent encountered, including ones not used in the final answer, plus inline citations as numbered links; X posts appear as x.com status links. The consumer apps show web and X sources, though how they display them is not formally documented. Shared conversations create public links that may be indexed by search engines.
What it means for you: Because Grok lists everything it saw, a citation is a weaker signal than elsewhere. Judge success by whether Grok names and describes you correctly in the answer itself, not only by whether your URL appears.
Crawlers
Grok's crawlers and user agents
The user-agent tokens to look for in your server logs and to name in robots.txt.
| User-agent token | What it does | robots.txt |
|---|---|---|
| Not published | SpaceXAI documents no web crawler or user-initiated fetcher for Grok. | Not published |
| GrokBot | Reported by a third-party directory, which classes it as an undocumented AI agent, with the user agent Mozilla/5.0 (compatible; GrokBot/1.0; +https://x.ai/). SpaceXAI has not said what it does. | Not published |
| Grok Bot agents Documentation | Grok Bot gives each AI 'teammate' a cloud computer with a browser that works across websites for users. No user agent is published, and traffic leaves through shared static IPs disclosed only to customers. | Not published |
Without a documented user agent or IP list, Grok-related traffic cannot be verified the way Googlebot, bingbot or Anthropic's bots can.
Crawl sources
Where Grok crawls from, and how to verify it
Not published. SpaceXAI publishes no IP ranges, reverse DNS hostnames or networks for any Grok crawler or search activity. For Grok Bot's hosted computers, it says traffic leaves through shared static egress IP addresses, given to customers by their account team only.
The sources you can influence are therefore the unnamed search engines behind Grok, through ordinary organic visibility, and X itself, whose public posts Grok retrieves directly from X's own data.
Opting out
How to opt out of Grok, or limit what it uses
robots.txt: nothing documented
SpaceXAI documents no robots.txt token. Rules for the third-party-reported GrokBot token may or may not be honoured. Grok's web results come from unnamed search providers, so their controls may matter more.
Official pageOpt out of training on X
On X, open Settings, then Privacy and safety, Data sharing and personalisation, and Grok and third-party collaborators, and untick the option allowing your public data and Grok interactions to be used for training. Feedback can still be used.
Official pageProtect your posts
Making an X account private keeps its posts out of Grok training and stops them being surfaced in Grok's answers.
Official pageControl grok.com and app data
Switch off 'Improve the model' under Data Controls, use Private Chat, which is deleted within 30 days and not used for training, or delete conversations.
Official pageRevoke shared links
Shared Grok conversations are public and may be indexed by search engines. You can revoke them at grok.com/share-links.
Official pagePrivacy requests
Use SpaceXAI's privacy portal for access, correction and deletion, or email its Data Protection Officer at privacy@x.ai. For Grok on X, use X's privacy form.
Official page
For brands, opting out of Grok is mostly self-defeating: protecting your X account removes you from Grok's answers entirely, and there is no documented way to stop it using the web. Grok-1 and Grok 2 weights already released keep what they learned.
How we help
How we get brands cited in Grok
The Coaley Peak GEO Research Centre researches and implements GEO for SEO agencies and in-house teams, with foreign markets as our speciality.
Grok visibility audit
We ask Grok your customers' questions and record what it says about you, which X posts and web pages it draws on, and which competitors it names.
An X presence that feeds answers
Public X posts are a first-party source for Grok. We help you post the facts, announcements and expert content that Grok can retrieve and repeat accurately.
Content informed by Grok's system prompts
xAI publishes Grok's system prompts. We read how Grok is told to search and cite, and structure your content to match.
Web visibility for Grok's search providers
Grok's web results come from engines SpaceXAI does not name. We build the organic visibility that makes your pages the ones they return.
Conversation monitoring
Because Grok repeats what people say on X, we monitor the conversation around your brand there and flag what may shape Grok's answers.
More on our research, services and team at the Coaley Peak GEO Research Centre.
FAQ
Grok GEO: questions
Does Grok use X posts in its answers?
Yes. Grok can decide to search public X posts as well as the web, and SpaceXAI says public X posts are provided by X. A brand's public posts, and what others post about it, can feed Grok's answers.
Does Grok have a crawler I can block?
SpaceXAI documents none. A 'GrokBot' user agent appears in a third-party directory, which classes it as undocumented. Grok's web results come from search providers SpaceXAI does not name.
How do I stop Grok training on my X posts?
Untick the Grok data-sharing option in X's privacy settings to stop your public data and Grok interactions being used for training. To stop posts being surfaced in Grok's answers as well, you would have to protect your account.
Who owns Grok?
Grok is made by SpaceXAI, formerly xAI, which has been a wholly owned subsidiary of SpaceX since February 2026. X Corp., which runs X, is a separate company in the same group.
Does Grok cite its sources?
Yes. The apps show web and X sources, and the API returns inline citations plus a list of every URL Grok encountered, including some not used in the final answer.
Are Grok's models open source?
Only older ones. Grok-1's base model was released under Apache 2.0 in March 2024, and Grok 2's weights under a community licence in August 2025. Grok 4.x models are closed.
Want a free chat about Grok?
Tell us which market and brand you are working on. If we can help, we will set up a free, no-obligation chat with Stephen or Ethan.
More engine guides
Related engines
Independence and trademarks. Grok is a trademark of SpaceXAI LLC (formerly xAI) or its affiliates; X is a trademark of X Corp. or its affiliates. Coaley Peak is independent: we are not associated with, endorsed by or a partner of SpaceXAI (formerly xAI). This guide is our own research from public documentation and reputable reporting, checked in October 2026.
Reviewed by Stephen Grindley, research centre lead. Last checked .
Document reference: ISO_webpage_geo_engine_grok_v1
Last modified: 7 October 2026