Microsoft Copilot and Grok visibility: how to show up in the two engines most GEO plans skip

Copilot visibility comes down to Bing: Microsoft documents that when Copilot searches the web, it turns the user's prompt into a short query, sends it to the Bing search service and builds its answer from what Bing returns. So to show up in Microsoft Copilot, your pages need to be in Bing's index, current, crawlable and free of the robots meta tags that keep content out of Bing's AI answers, and you can watch the result in Bing Webmaster Tools' AI Performance report. Grok visibility is a different problem. xAI documents web and X search tools for Grok but publishes no crawler you can allow or block by name and no reporting for site owners, so the levers are indirect: be readable to general crawling, be corroborated on the directories and communities Grok leans on, keep your facts consistent on X, and measure Grok's answers yourself, counting only the sources it actually cites.

This guide separates what Microsoft and xAI document from what they do not, because in both engines the undocumented part is where most bad advice lives. Where we have first-party measurements (for Grok, across 12 industry benchmarks) we show them with their caveats. Where we do not (we do not currently measure Copilot answers), we say so and stick to what Microsoft has published.

Key takeaways

• Copilot grounds web answers through the Bing search service. Getting into and staying fresh in Bing's index is the foundation of Copilot visibility; IndexNow, sitemaps and Bing Webmaster Tools are the tools.
• Bing's published robots meta rules matter: in its 2023 rules for Bing Chat, NOARCHIVE kept content out of answers entirely, and NOCACHE limited it to URL, title and snippet.
• Bing Webmaster Tools now reports citations in Copilot and Bing's AI summaries, including the grounding queries that retrieved your pages. It is a preview and a sample, but it is the only first-party AI citation report any major engine gives site owners.
• Grok has no documented crawler user agent, so you cannot allow or block it by name. Do not invent one in robots.txt.
• In our benchmarks, Grok attached more citations per answer than ChatGPT, Claude and Gemini in 11 of 12 industries, and was the engine most likely to send readers to directories and review sites in several local categories.

Why Copilot visibility and Grok visibility get skipped

Most GEO plans start with ChatGPT, Google's AI features and Perplexity, and stop there. There are understandable reasons. Neither Copilot nor Grok has a named crawler in the way OpenAI, Anthropic and Perplexity do. Few monitoring tools covered them until recently. And both are perceived as niche next to ChatGPT.

The perception is the weakest of those reasons. Copilot sits inside Windows, Edge and Microsoft's workplace apps, so for many office workers it is the assistant that is already open when a work question comes up. Grok sits inside X, where fast-moving narratives about brands, products and incidents play out. If your buyers work in Microsoft-heavy organizations, or your category is discussed on X, skipping either engine leaves a blind spot exactly where some of your audience asks questions.

The good news is that the work is mostly not new. Copilot rewards the same index hygiene that Bing has always rewarded, plus a few AI-specific controls. Grok rewards the same readability and corroboration that every answer engine rewards, with a different weighting. What changes is the toolkit and the measurement.

Which of the two to prioritise

Both engines deserve a baseline check, but the depth of work should follow your audience. A few rules of thumb:

• Prioritise Copilot if you sell to businesses whose staff work in Microsoft's workplace apps, if your buyers research during the working day on managed devices, or if you already get meaningful traffic from Bing. Copilot work also pays back in Bing search itself, so it is rarely wasted.
• Prioritise Grok if your category has an active conversation on X (check by searching your category and competitors there), if your brand is the subject of fast-moving news, or if you are a local or consumer business whose reputation lives on review platforms.
• Do both lightly if neither applies strongly: the Bing index checks in this guide take a day, and a monthly sample of Grok answers on your core questions takes an hour.

Whichever you choose, write the decision down with the reason. Six months later, when someone asks why the team is or is not working on an engine, the reason is more useful than the decision.

What we can and cannot show you

We measure Grok as one of five engines in our industry benchmarks, so the Grok sections include real data from 240 Grok answers collected on 2026-09-07. We do not currently measure Copilot answers. The Copilot sections therefore rest on Microsoft's own documentation and announcements, and we flag where a widely repeated claim could not be verified against a Microsoft source.

How Microsoft Copilot finds and cites your pages

Microsoft's documentation on data, privacy and security for web search in Microsoft Copilot and Copilot Chat is unusually specific about the mechanics, and every step has a consequence for site owners.

Step by step, with what it means for you

1. Copilot decides whether the web would help. When web search is enabled, Copilot parses the prompt and identifies terms where web information would improve the answer. Not every prompt triggers a web search; questions answerable from the model or from a user's own files may not.
2. It writes its own short query. Microsoft says the generated query is different from the user's prompt and consists of a few words informed by it. A long, conversational prompt about "the best way to automate invoice approvals for a 40-person team" may become a short query that looks much more like a traditional search. That means classic search relevance for short queries still matters for Copilot: the page that answers "invoice approval automation" clearly is the page Bing returns.
3. The query goes to Bing. Microsoft says the generated query is sent to the Bing search service without the user's entire prompt (unless it is very short), without entire files, and without identifying information such as username, domain or tenant ID.
4. Bing returns results from its index. This is the step you influence most. A page Bing has not crawled, or has crawled in an old version, cannot ground the answer.
5. Copilot composes the answer. The response is built from the returned information, with citations. In Copilot Chat, Microsoft says the exact web search queries are shown to users in the linked citation section, where they remain available in the chat thread for 24 hours.

Two details that change how you test

First, the query transparency in Copilot Chat is a free research tool. When you test a question in Copilot Chat, open the citation section and read the exact queries it sent to Bing. Then search those queries in Bing yourself. If your page is not in the top results for the generated query, the problem is Bing relevance for that query, not anything specific to Copilot.

Second, web search in Copilot can be switched off. Microsoft documents an admin policy, "Allow web search in Copilot", and a user-level Web content toggle. In organizations that disable web search, Copilot answers work questions without the web at all, and no amount of GEO reaches those answers. When you test Copilot from a work account, check whether web search is enabled before drawing conclusions. Microsoft also says the generated queries do not affect Bing's search ranking, so testing Copilot does not distort your own Bing results.

Testing a question in Copilot, step by step

1. Pick a question from your question set and write it the way a customer would, including context (company size, location, constraint).
2. Ask it in Copilot with web search confirmed on. Save the full answer and the list of cited sources with the date.
3. In Copilot Chat, open the citation section and copy the exact web queries Copilot generated. Note how much shorter and more search-like they are than your question.
4. Run each generated query in Bing. Record where your best page ranks, and which pages outrank it.
5. Compare the pages Copilot cited with Bing's top results for the generated queries. If they overlap heavily, your lever is Bing relevance for those queries; if Copilot cited pages lower down, it is choosing for content, so your lever is the clarity of the page itself.
6. Repeat the same question next month. Generated queries can differ between runs, so keep every version you see; a set of query variants is more useful than any single one.

What Microsoft documents, and what it does not

The second row deserves emphasis. Lists circulating online sometimes include a "Copilot bot" user agent. We could not verify one in Microsoft's public webmaster documentation, which is why our AI crawler list leaves Copilot without a row rather than guessing. A robots.txt rule written for a guessed name does nothing, and gives you false confidence that you have allowed or blocked something.

How to show up in Microsoft Copilot: start in the Bing index

If Copilot grounds through Bing, Copilot visibility starts with Bing visibility. Many sites have spent years optimizing for Google and have never looked at how Bing sees them. Start there.

Step 1: verify the site in Bing Webmaster Tools

Verify ownership, submit your XML sitemaps, and read the crawl and indexing reports. Look for three things in particular: pages you care about that are not indexed, pages indexed with an old version, and crawl errors or blocks.

Step 2: make sure Bingbot is allowed and can render the page

Check robots.txt for rules that block Bingbot, explicitly or through a broad User-agent: * disallow. Check your CDN or bot-management rules, which sometimes challenge crawlers that are not Googlebot. Then check rendering: if your key content only appears after client-side script runs, test what Bing's crawler actually receives. Our guide to JavaScript rendering and AI covers the checks; the rule of thumb is that the facts you want quoted should be in the HTML the server sends.

Step 3: tell Bing about changes with IndexNow

IndexNow is a protocol for telling participating search engines that a URL has changed, instead of waiting for them to recrawl. Bing's own AI Performance announcement recommends it to keep information fresh across search and AI experiences. The IndexNow documentation sets out the mechanics: generate a key of 8 to 128 hexadecimal characters, host it as a text file at the root of your site, and submit changed URLs, either one at a time or up to 10,000 per POST request. Submitted URLs are shared with all other participating engines. The IndexNow FAQ lists Amazon, Bing, Naver, Seznam.cz, Yandex and Yep as participants; Google is not among them, so IndexNow complements your Google setup rather than replacing it.

# 1. Host your key at the site root (the key is your own hex string)
https://www.example.com/0f3c9a7e5b2d4c18a6e7.txt   ->  file contents: 0f3c9a7e5b2d4c18a6e7

# 2a. One changed URL
GET https://api.indexnow.org/indexnow?url=https://www.example.com/pricing/&key=0f3c9a7e5b2d4c18a6e7

# 2b. A batch of changed URLs
POST https://api.indexnow.org/indexnow
Content-Type: application/json; charset=utf-8

{
  "host": "www.example.com",
  "key": "0f3c9a7e5b2d4c18a6e7",
  "keyLocation": "https://www.example.com/0f3c9a7e5b2d4c18a6e7.txt",
  "urlList": [
    "https://www.example.com/pricing/",
    "https://www.example.com/help/billing/"
  ]
}

Most CMSs and CDNs have IndexNow plugins or built-in options, so you rarely need to write this yourself. What matters is the trigger: submit on publish, on meaningful update and on deletion, and do not resubmit unchanged URLs on a timer.

Step 4: keep sitemaps honest

Sitemaps remain the other half of discovery. Keep them limited to canonical, indexable URLs, and keep lastmod truthful: update it when the content meaningfully changes, not on every deploy. A sitemap where every URL claims to have changed today teaches crawlers to ignore the field. Our post on sitemaps and AI search goes further.

Step 5: match short queries with clear pages

Because Copilot sends a short generated query, the page that wins is the one that answers that short query clearly. In practice:

• Give each important question its own page or clearly headed section, with the answer in the first paragraph under the heading.
• Use the terms people would type into a search box in titles and headings, alongside the conversational phrasing.
• Put facts in structures Bing recommends for AI reference: clear headings, tables and FAQ sections, as Bing's AI Performance announcement puts it.
• Support claims with examples, data and sources, which Bing also recommends, and keep text, images and video consistent about the same entities.
• For local businesses, keep Bing Places for Business details current; Bing's announcement singles this out for local visibility.

Step 6: control how Copilot uses your content

Bing has published robots meta controls specifically for its AI answers. On September 22, 2023, Bing set out how NOCACHE and NOARCHIVE work for Bing Chat, the product that has since become Copilot.

Step 7: audit the NOARCHIVE and NOCACHE you already have

Many sites carry noarchive from an old decision about cached copies in search results, unrelated to AI. Under Bing's rules for Bing Chat, that tag keeps the page out of AI answers entirely. Search your templates and HTTP headers for noarchive and nocache, in both the robots meta tag and the X-Robots-Tag header, and confirm each one is intentional. Our guide to meta robots and X-Robots-Tag explains where these directives live and how they combine.

A caveat on how current these rules are

These rules were published in 2023 for Bing Chat. Microsoft has since renamed and expanded Copilot, and third-party reports describe updated Bing Webmaster Guidelines covering Copilot. We could not read the current guidelines page directly to verify the details, so treat the table as Bing's last rules we could verify and check Bing's current Webmaster Guidelines before changing tags on pages that matter.

Measuring Copilot visibility with Bing Webmaster Tools AI Performance

On February 10, 2026, Bing introduced AI Performance in Bing Webmaster Tools as a public preview. It reports citation activity across Microsoft Copilot, AI-generated summaries in Bing and select partner integrations. For Copilot visibility, it is the closest thing to Search Console for AI that any major engine offers.

The June additions arrived on June 16, 2026, when Bing began rolling out Intents, Topics, Citation Share and Compare in preview. Bing is explicit about the limits of Citation Share: it does not expose competitor domains, represent traffic share, or assign quality scores to content. Read it as "how much of the cited space on this query is mine", nothing more.

A monthly loop

1. Read. Export grounding queries and page-level citations for the last period. Group the queries by intent and topic.
2. Map. For each query group, name the page on your site that should answer it. Where the cited page is the wrong one (an old blog post instead of the current product page), or where no page of yours is cited, note it.
3. Improve and ping. Fix the page that should answer: clear heading, direct answer, current facts, tables where they help. Submit the URL through IndexNow.
4. Compare. Next period, use Compare to overlay the previous period and check whether the query group moved to the right page and whether citations rose.

Reading a grounding query report: what to look for

The first export is usually surprising. Four patterns come up often enough to look for deliberately:

• The wrong page is cited. A query about pricing is answered from a three-year-old blog post that mentions prices, rather than from the pricing page. Update or redirect the old post, make the pricing page answer the query in its first paragraph, and submit both through IndexNow.
• Only definitional queries reach you. Intents show that you are cited for "what is" questions but not for commercial ones. Your explainers are working; your product and comparison pages are not being retrieved for buying queries. Look at how those pages are titled and whether they answer a question at all.
• One page carries everything. Average cited pages is low and one URL dominates. That page is a single point of failure; build supporting pages for the sub-questions it is being cited for.
• Queries you never targeted. Grounding queries show topics you did not know you were relevant for. Some are opportunities worth a dedicated page; some are mismatches worth clarifying so you are not cited for something you do not do.

What the report cannot tell you

Citations are not clicks or visits. The report shows when your content was a source, not whether anyone followed the link. Grounding queries are shown as a sample, and the newer features are in preview, so small numbers will be noisy. And the report covers Microsoft's surfaces only. Use it alongside your own measurement of answers on the questions that matter to you, not as a replacement for it.

How Grok retrieves and cites sources

xAI's developer documentation describes the retrieval tools available to Grok through its API. The consumer Grok experience is not documented in the same detail, so read this section as the best public evidence of how Grok can look things up, not as a full description of the app.

Two search tools

The Web Search tool lets Grok search the web in real time and browse web pages. Developers can restrict it with allowed_domains or exclude sites with excluded_domains, each limited to 5 domains and not usable together. The X Search tool lets Grok perform keyword search, semantic search, user search and thread fetch on X, with optional filters for up to 20 included or excluded handles and a date range.

Two implications follow. First, X is a first-class source for Grok in a way it is not for other engines. For topics where conversation on X moves fast (launches, outages, controversies, trends), what is said on X about your brand can enter Grok's answers directly. Second, applications built on the Grok API can narrow or exclude sources. An assistant built by a developer on Grok may only look at a handful of domains, which you cannot influence except by being on the sites developers in your space trust.

Two kinds of citation

xAI's citations documentation distinguishes a complete list of all sources encountered during search, returned by default, from inline citations embedded in the answer text. xAI says not every URL in the full list is necessarily referenced in the final answer, and that the model decides when and where to place inline citations, so they are not guaranteed on every answer. For measurement this matters a great deal. If you count the full encountered list as "citations", you count every page Grok glanced at, and a domain will look far more cited than it is. Count inline citations.

What is not documented

xAI does not publish a documented crawler user agent. You cannot allow or block Grok by name in robots.txt, and any rule written for a guessed name, such as a "GrokBot" group, is a rule for an agent that may not exist. What you control is whether your pages are readable to general crawling at all: real HTML, and no bot-management rule that returns errors or challenges to unknown agents fetching public pages. Our Grok engine page keeps the current state of what is and is not documented.

Grok visibility in practice: what our benchmarks show

We measure Grok as one of five engines across 12 industry benchmarks, each with 20 buyer questions, collected on 2026-09-07. That gives 240 Grok answers, and every one of the 12 industries had citations on all 20 of its Grok answers.

A method caveat first

In these benchmarks, every engine except Perplexity was queried as a model through an API with the same web-search tool attached, and only sources marked inline in the answer text were counted. For Grok that means the answers reflect Grok's own choices about what to say and what to cite, but the pool of search results came from that shared search layer, not from xAI's own Web Search or X Search tools. That is why no answer in our data cites X, and why the consumer Grok app, which can search X, may cite differently. Read the numbers below as Grok's answering behaviour, not as a map of where the Grok app looks.

Grok cites more than most engines

Across the 12 industries, Grok attached between 9.35 and 11.7 citations per answer on average. That was more than ChatGPT, Claude and Gemini in 11 of the 12 industries (ChatGPT edged ahead only in law firms), and less than Perplexity in all 12. A citation-heavy engine is good news for sites that are well represented in search results for their topics: there are more slots to land in. It also means that being cited once by Grok is a weaker signal than being cited once by a sparse engine.

Grok leans on directories and review sites

In local and consumer categories, Grok frequently pointed readers to directories and review sites rather than, or as well as, naming providers. It did so in 10 of 20 answers about dentists in Austin and hotels in Miami, and 9 of 20 about restaurants. In the software category it never did. Compared with the other engines on the same questions, Grok was often the most directory-leaning.

In the Miami hotels benchmark, Grok named a directory or review site in 10 of 20 answers, against 8 for ChatGPT, 7 for Claude, 4 for Perplexity and 3 for Gemini. For a local or consumer business, this makes directory and review profiles more important for Grok visibility than for most engines: if Grok sends the reader to a review platform, your profile there is effectively your answer.

What we would not conclude

One collection day, 20 questions per category, and a shared search layer. These numbers describe how the Grok model answered in that setup. They do not tell you how often the Grok app is used for your category, how its X search changes results, or how answers move over time. Treat them as a reason to check Grok in your own category, not as a forecast.

A Grok visibility playbook

With no crawler to allow, no content-use controls and no first-party reporting, Grok visibility work is about making sure that wherever Grok looks, it finds an accurate picture of you.

1. Be readable to general crawling

Since Grok cannot be allowed by name, your User-agent: * rules and your bot-management settings decide what any undocumented agent can fetch. Make sure public pages (product, pricing, help, about) return real HTML with the important facts in it, and that your firewall or CDN does not answer unknown agents on public pages with a 403 or a challenge. If you deliberately block unknown agents for security reasons, know that you are accepting a visibility cost in engines without named crawlers.

# There is no documented Grok crawler, so there is no Grok group to write.
# Whatever this group allows is what undocumented agents can read.
User-agent: *
Allow: /
Disallow: /account/
Disallow: /checkout/

# Named groups only for agents whose operators document them, e.g.:
User-agent: Bingbot
Allow: /
Disallow: /account/
Disallow: /checkout/

Sitemap: https://www.example.com/sitemap.xml

2. Fix the directory and review layer

The benchmarks show Grok leaning on directories and review sites in local and consumer categories. List the platforms that appear in Grok's answers for your questions, then make your profiles there complete, current and consistent with your site: name, categories, locations, hours, prices where shown, and a steady flow of genuine reviews collected under uniform rules.

3. Treat X as a source, not just a channel

Because Grok can search X directly, your presence there is part of your evidence base. Keep an official account with a clear bio that matches your site's description of what you do, pin accurate statements about launches, pricing changes and incidents, and correct misinformation with links to your own pages. Do not manufacture engagement or run undisclosed accounts; besides the obvious platform and regulatory risks, it creates exactly the kind of inconsistent picture that answer engines discount.

Once a month, search X for your brand name, your product names and the questions your customers ask, sorted by recent. Note recurring claims about you, accurate or not, and which accounts make them. Where a claim is wrong, the fix is a clear, dated statement on your own site and a reply that links to it, not an argument. Where a claim keeps recurring because your own documentation is unclear, fix the documentation; that is the source Grok's web search will find next to the posts.

4. Keep facts consistent across the open web

Grok, like every engine, is more confident about facts it can corroborate. Your about page, company profiles, press and directory entries should agree on what you are, who you serve and what you cost. Where they disagree, the answer will hedge or pick the wrong one.

5. Publish answers for fast-moving questions

Grok's tools are built for current information. When something changes in your business that people will ask about (a new product, a price change, an outage, a policy change), publish a dated, plain statement on your site promptly and link to it from X. Old information that is never superseded tends to stay in circulation.

6. When Grok or Copilot disagrees with the other engines

It is common to be named by ChatGPT and Gemini and missing from Copilot or Grok, or the reverse. Diagnose before you react. If Copilot misses you while Google-grounded engines include you, compare your Bing and Google indexing for the cited pages; differences between the two indexes are the usual cause, and Bing Webmaster Tools will show them. If Grok misses you while others include you, look at the directories and review sites Grok cites for the question; if you are absent or thin there, that is the gap. If Grok includes you and others do not, check that what it says is accurate before celebrating, because a directory-led answer can carry outdated details from a stale profile.

Measuring Copilot and Grok visibility side by side

The two engines need different measurement setups, because one offers first-party data and the other offers none.

Building the question set

Use the same question set for both engines, so the comparison is fair: buyer questions in customers' words, split by intent (finding a provider, comparing options, checking trust, learning how something works). Keep questions that name your own brand out of the visibility rate and report them separately. For Grok, add a few questions about recent events in your category, since that is where its X search makes it behave differently from other engines.

Reading the results

For each engine, record whether you were named, whether you were cited, which page was cited, and which other domains were cited. Compare like with like over time: the same questions, on the same engine, monthly. Report an engine you did not measure as "not measured", never as zero.

A 30-day plan for Copilot and Grok visibility

Week 1: access

• Verify the site in Bing Webmaster Tools, submit sitemaps, and fix crawl and indexing errors on important pages.
• Check robots.txt, CDN and bot-management rules for anything that blocks Bingbot or challenges unknown agents on public pages.
• Search templates and headers for noarchive and nocache; keep only the intentional ones.

Week 2: freshness

• Set up IndexNow through your CMS or CDN, triggered on publish, update and deletion.
• Clean sitemaps so lastmod reflects real content changes.
• Audit directory and review profiles that matter in your category for accuracy and consistency.

Week 3: content

• For your most important questions, make sure one page answers each clearly, with headings, a direct first paragraph, tables where they help, and sources.
• Align your X profile and pinned statements with your site's description and current facts.

Week 4: measurement

• Read AI Performance: grounding queries, cited pages, intents. Map query groups to pages.
• Run your question set in Copilot and Grok; record names, inline citations and cited pages.
• Set the monthly loop: AI Performance review, IndexNow on every fix, and the same question set each month.

If you want to see which domains each engine cites for your questions without running the sampling by hand, our citation sources report does that for the engines we measure.

Sources

• Microsoft Learn, "Data, privacy, and security for web search in Microsoft Copilot and Microsoft Copilot Chat" (updated 2026-08-18): learn.microsoft.com/en-us/microsoft-365/copilot/manage-public-web-access
• Microsoft Bing Webmaster Blog, "Introducing AI Performance in Bing Webmaster Tools Public Preview" (2026-02-10): blogs.bing.com
• Microsoft Bing Search Blog, "New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare" (2026-06-16): blogs.bing.com
• Microsoft Bing Webmaster Blog, "Announcing new options for webmasters to control usage of their content in Bing Chat" (2023-09-22): blogs.bing.com
• IndexNow documentation and FAQ: indexnow.org/documentation, indexnow.org/faq
• xAI documentation, Web Search, X Search and Citations: docs.x.ai/developers/tools/web-search, docs.x.ai/developers/tools/x-search, docs.x.ai/developers/tools/citations
• InsightWonder industry benchmarks, 12 categories × 5 engines × 20 questions, collected 2026-09-07; Grok engine page: insightwonder.com/ai-engines/grok/