Grok SEO: the technical guide to getting cited by Grok

How Grok searches the web and X#

Grok answers from training when it can and searches when a question needs something current. SpaceX’s May 2026 S-1 describes Grok Chat as offering “real-time integration of web search, X data, code execution, and multimodal analysis” on grok.com, the Grok apps, inside X and through the API. xAI — acquired by SpaceX effective 2 February 2026 and now branded SpaceXAI — gives its API the same two search tools, and their documentation is the clearest public description of what Grok can do.

The Web Search tool lets Grok “search the web in real-time and browse web pages.” The X Search tool performs “keyword search, semantic search, user search, and thread fetch on X.” What the docs leave out matters as much: xAI names no search provider, publishes no index size and documents no crawler. Where Grok’s web results come from is not public.

1. 1

   Grok decides whether to search

   The model “decides what to do next: make a tool call, or provide a final answer,” per xAI’s tools overview, and can chain several tools per question. Stable knowledge comes from pre-training, which the Grok 4 model card says includes “publicly available Internet data.”

   Your lever: Be in the training data: public, crawlable pages are the only route, since xAI gives websites no documented opt-in or opt-out token.

2. 2

   It searches the web and X, often in parallel

   xAI’s published prompt for @grok says “parallel search should be used to find diverse viewpoints.” The chat prompt tells Grok to “search for a distribution of sources that represents all parties/stakeholders” on controversial questions, and not to “shy away from deeper and wider searches” on X.

   Your lever: Be one of the viewpoints it finds: a first-party page for the facts, and posts on X that state them.

3. 3

   An undisclosed web index returns candidates

   xAI doesn’t name the index or say how it ranks results. There’s no webmaster console, no URL submission and no IndexNow participation.

   Your lever: Keep your own pages crawlable, and earn mentions where Grok’s citations already concentrate: Reddit, YouTube, review sites.

4. 4

   Grok opens the pages it needs

   Grok 4 Fast “hops through links, ingests media (including images and videos on X),” per xAI’s September 2025 launch post. The fetches carry browser-like user agents from datacenter IPs, and in a June 2026 test the answer came from raw HTML.

   Your lever: Server-render, and don’t put a bot challenge in front of content pages.

5. 5

   It answers with inline citations

   The API inserts numbered links like [[1]](https://x.ai/news). grok.com renders citations as small pills showing the site’s name or an X @handle, and an “N sources” pill under the answer opens the full list. Presenc AI counted an average of 4.7 sources per answer in 2026.

   Your lever: Write facts that stand on their own, on pages whose titles read well on a source card.

Crawlers and access: nothing to allow by name#

OpenAI, Anthropic, Google and Perplexity all publish their AI crawlers. xAI doesn’t: there’s no crawler page, no user-agent token, no IP list and no statement on robots.txt for Grok’s search or live fetches. Cloudflare wrote in September 2025 that “xAI’s bot, grok, does not self-identify at all, making it impossible for website operators to block it.” A February 2026 controlled test by StackFox found the same: asked to fetch test URLs, Grok sent 30 requests in under a second from datacenter IPs, with user agents like Chrome/139.0.0.0 and Go-http-client/1.1. “The word ‘Grok’ never appears. Neither does ‘xAI.’”

The consequence: robots.txt can’t target Grok, and neither can a CDN’s “block AI bots” switch, because nothing in the request says it’s Grok. Nothing documents whether Grok’s fetcher reads robots.txt at all. Adding the unverified tokens does no harm, but don’t mistake them for control:

# xAI documents no user agent for Grok's search or fetches.
# These names come from third-party lists and are unverified:
# a rule for them may never match a real request.
User-agent: GrokBot
User-agent: xAI-Grok
User-agent: Grok-DeepSearch
Allow: /
 
# Grok's browser-like fetches fall under your default group
User-agent: *
Allow: /
Disallow: /account/

Training: no website opt-out, one X setting

The Grok 4 model card lists “publicly available Internet data” among Grok’s pre-training sources. With no documented crawler token, site owners have no robots.txt line that xAI says it honors for training. X posts are different: X’s privacy policy, effective 15 January 2026, says X may use what it collects “and publicly available information to help train our machine learning or artificial intelligence models.” Account holders can opt out under Settings and privacy → Privacy and safety → Grok & Third-party Collaborators, which PCMag describes as covering public posts, engagements and profile data. As far as xAI documents, that setting governs training, not whether X search can find and cite your public posts.

X is Grok's second index#

Grok’s second source is X itself. Its X search runs keyword and semantic search over posts, looks up accounts and fetches whole threads, with filters for handles and date ranges and optional image and video understanding. Cited posts come back as x.com/<handle>/status/<id> URLs, per xAI’s citations docs, and grok.com shows them as @handle pills and profile photos in the sources stack.

How much of Grok’s citing goes to X depends on who’s counting. Presenc AI sampled 4,400 prompts in March–April 2026 and found about 45% of Grok’s citations were X posts: “The mix shifts toward X heavily for current-events queries and toward web for definitional or how-to queries.” Ahrefs’ Brand Radar, across 1.9 million broad US queries, put x.com at 1.4% of the citations among Grok’s top 50 domains. Both can be right — a product question gets answered from the web, a what’s-happening question from X — so plan for both.

• State the fact in the post’s text. X search matches keywords and meaning. Grok can read images and video, but a claim that lives only in a screenshot or a link preview is the harder one to find.
• Link the post to the canonical page. Grok’s system prompt says it can “analyze individual X user profiles, X posts and their links” — a post is a route to your page.
• Answer recurring questions in threads. Thread fetch reads a whole conversation in one call, so a thread that settles a question completely is a ready-made source.
• Make the account unambiguous. User search is a documented capability. Put the brand name in the handle or display name, and say what you do in the bio.
• Consider verification for the brand account. Presenc AI found verified accounts’ posts cited at 2.8x the rate of equally-engaged unverified ones.
• Post on X the day your facts change. The X tool filters by date and Grok leans on X for current events. A price change or launch that exists only on your blog is easier to miss.

Grokipedia: worth checking, not a Grok lever#

Grokipedia launched on 27 October 2025 with 885,279 articles and passed 6 million by January 2026, per its own article. Some articles are generated by Grok and others were forked from Wikipedia. Nobody edits directly: logged-in users highlight text, click Suggest Edit and submit a correction with supporting evidence, which Grok reviews before anything changes.

Its documented reach has been outside Grok. In January 2026 TechCrunch reported the Guardian’s finding that GPT-5.2 cited Grokipedia nine times across more than a dozen questions. Ahrefs’ March 2026 comparison counted 356,200 AI citations for Grokipedia against 24.9 million for Wikipedia, and found its organic traffic had fallen to half its peak. For Grok itself, xAI doesn’t document any use of Grokipedia in answers, and grokipedia.com isn’t among the 50 domains Grok cites most in Ahrefs’ June 2026 data.

• Search Grokipedia for your brand, products and founders. Generated pages can exist without anyone having asked for them.
• Correct what’s wrong, with a primary source. Link your own documentation, a filing or a dated announcement in the suggestion form.
• Don’t build a Grok strategy on it. Its documented citations are in ChatGPT and other assistants, not in Grok’s top cited domains.

Technical requirements#

Nothing about Grok’s fetcher is documented — not its user agent, rendering, timeouts or size limits. What’s known comes from tests, and it points the same way as every other engine: plain HTML, reachable without a challenge.

Server-rendered HTMLRequired

In Search Engine World’s June 2026 test, one Grok fetch node ran the page’s JavaScript, but the answer still quoted the raw-HTML decoy. Put the content in the HTML response — see server-side rendering.

No challenge on content pagesRequired

Grok’s fetches arrive as browser bursts from datacenter IPs. A JavaScript challenge, CAPTCHA or datacenter block hands them a page with none of your content.

Tolerant rate limitsHelps

StackFox logged 30 requests in under a second when Grok fetched its test pages. Per-second limits tuned for human visitors will cut some of them off.

Descriptive titlesHelps

grok.com’s source cards show the page title in bold over a three-line preview, and inline pills show the site’s display name. The title is what a reader sees before clicking.

Links from X postsHelps

Grok can follow the links in the X posts it reads, so posts that link your canonical URL are a second path to the page.

Structured dataUnconfirmed

xAI has said nothing about schema.org. Keep every fact that matters in visible text as well.

SitemapsUnconfirmed

Not documented for Grok. Keep one with honest lastmod values for the engines that do use them.

Canonicals and redirectsUnconfirmed

Undocumented. Keep one clean URL per page and avoid redirect chains, so whichever copy Grok finds is the right one.

llms.txtUnconfirmed

xAI hasn’t said Grok reads llms.txt, and no test has shown it requested.

robots.txt rules naming GrokNo effect

There’s no documented token to name. A rule for GrokBot or xAI-Grok may never match a request.

IndexNowNo effect

xAI isn’t on IndexNow’s list of participating engines, so pings don’t reach Grok.

Test pages the way Grok’s fetcher appears to see them — a browser user agent, no JavaScript — ideally from a cloud VM, since that’s where its requests come from:

# A browser user agent and no JavaScript, like Grok's fetches
curl -s -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/139.0.0.0 Safari/537.36" \
  https://yoursite.com/pricing | grep -c "Plans start at"
 
# 0 = the text needs JavaScript, or your WAF served a challenge page

What Grok cites#

Ahrefs’ June 2026 ranking of the 50 domains Grok cites most, built from 1.9 million US queries, is dominated by user-generated platforms: Reddit 16.3%, YouTube 15.1% and Facebook 13.9% — about 45% between them — then Instagram, Quora, Amazon, TikTok and Wikipedia. Retail and review sites fill much of the rest: eBay, Walmart, Consumer Reports, Yelp, Tripadvisor and Trustpilot.

• Community platforms

  Observed

  Reddit, YouTube and Facebook are Grok’s top three cited domains. Being discussed there is a route in that doesn’t depend on your own rankings — see Reddit SEO.

• Marketplaces and reviews

  Observed

  Amazon (5.0%), eBay, Walmart, Consumer Reports and Trustpilot all make Grok’s top 50. Product questions get answered from listings and reviews.

• X posts, by query type

  Observed

  About 45% of citations in Presenc AI’s sample, 1.4% in Ahrefs’. The share is heaviest for current events and lightest for how-to questions.

• Verified accounts

  Observed

  Posts from verified accounts were cited at 2.8x the rate of equally-engaged unverified posts in Presenc AI’s 2026 sample.

• Primary sources

  Official

  @grok is told to “fetch primary sources for current events” and to verify search results by opening pages. Your own announcement page is the primary source for your news.

• A spread of viewpoints

  Official

  On contested questions Grok is told to search for “all parties/stakeholders” and to treat media viewpoints as biased. Expect your page to be cited next to others, not alone.

Paid placement isn’t a route yet. Elon Musk told advertisers in August 2025 that X planned ads in Grok’s responses. SpaceX’s May 2026 S-1 still described advertising in the stand-alone Grok product as a possible “incremental monetization opportunity,” and said it doesn’t sell advertisers placement on the Grok API.

Measuring Grok traffic#

Expect little Grok referral traffic in your analytics — and some of the shortfall is measurement. SE Ranking’s study of 101,574 sites found Grok’s share of all website traffic peaked at 0.0019% in August 2025 before falling back toward zero. Goodie’s B2B panel saw “effectively zero attributable referrals” from Grok while Similarweb counted 904 million visits to it in January–April 2026.

Part of that is how grok.com links out. Its web client, as shipped on 21 September 2026, renders citation pills and source cards with rel="noopener noreferrer nofollow" and adds no UTM tags, so the browser sends no referrer and those visits land in Direct. Links without that attribute inherit the site’s Referrer-Policy: origin-when-cross-origin and arrive as grok.com. GA4’s AI Assistant channel, added in May 2026, names Grok — Google’s channel documentation lists “ChatGPT, Gemini, Deepseek, Copilot, or Grok” — but it can only classify visits that carry a referrer. How clicks from Grok inside the X app are attributed isn’t documented.

# Session source — Grok's web client, when a referrer survives
^(.*\.)?grok\.com$
 
# X itself: t.co is the link wrapper on X posts
^(t\.co|x\.com)$

Server logs are the better instrument, with one catch: you can’t filter Grok by user agent. Two things work instead. Look for its signature — bursts of requests for one URL within a second, browser user agents, no referrer — and run a controlled test: publish a URL nobody knows, ask Grok to read it, and see exactly what arrives.

# Bursts: 10+ no-referrer requests for one URL in one second (combined log format)
awk '$11=="\"-\"" {print substr($4,2,20), $7}' access.log \
  | sort | uniq -c | awk '$1 >= 10' | sort -rn | head -20
 
# Controlled test: ask Grok to summarize https://yoursite.com/grok-test-7f3a, then
grep "/grok-test-7f3a" access.log

For visibility itself, run a fixed prompt panel through Grok on a schedule. Mix current-events prompts, where X dominates, with product and how-to prompts, where the web does, and record which domains and @handles get cited. Answers vary between runs, so compare across repeats rather than trusting one screenshot. Rankbox’s free AI visibility prompt kit gives you a starting set.

Myths worth dropping#

The action checklist#

Everything above, in the order we’d do it. Tick items off as you go — your progress is saved in this browser.

Grok SEO: frequently asked questions#