How do I get found in Microsoft Copilot?

Someone on your team ran a Copilot query about vendors in your category, watched it name two companies that were not yours, and forwarded you the screenshot. So you went looking for the Copilot equivalent of GPTBot — a user agent to allow, a documentation page to read, a file to add. There is not one.

Microsoft does not operate a Copilot crawler. That single fact turns a vague question about AI visibility into an ordinary technical SEO job, and the rest of this post is that job.

There is no Copilot crawler, and Microsoft says so

Microsoft's Copilot Studio documentation on generative answers from public websites lays out the plumbing directly. The crawler is Bingbot. The retrieval service is Bing Custom Search. The page's own line is plain: “As Bingbot crawls the web, it sends information to Bing about what it finds.” Microsoft then instructs site owners to use robots.txt to tell Bingbot which pages it can and cannot access — a strange instruction to give if some separate Copilot agent were doing the fetching.

So the question collapses. “How do I get found in Copilot” is “am I properly indexed in Bing,” and that is a question most marketing teams have not asked seriously since roughly 2014, because Bing's organic share never justified the attention. The neglect is the opportunity. There is a good chance nobody on your competitor's team has looked at their Bing index coverage this year either.

# Bing is the retrieval layer behind Microsoft Copilot.
# Allowing Bingbot IS the "Copilot crawler" configuration.

User-agent: bingbot
Allow: /

User-agent: msnbot
Allow: /

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

The pipeline between the question and the answer

Microsoft documents what happens in between, and each stage tells you something about what to fix.

1. Message moderation. Copilot parses the query and filters malicious content. Nothing here is yours to influence.
2. Query optimization. It adds context from the conversation history, including location and time information.
3. Information retrieval. The rewritten question is converted into a Bing Custom Search query. If you are not in Bing's index, your involvement ends at this line.
4. Collation and parsing. Copilot collates and parses the relevant top results — the top of the result set, not the whole of it.
5. Grounding, provenance and semantic similarity checks. Retrieved passages are tested before anything is written.
6. Summarization. What survives is rewritten into plain language and delivered to the user.

Two stages deserve more than a line. Query optimization means the string Bing searches is not the string the user typed, so you are competing on rewritten queries you never see. The closest published evidence for what that rewards comes from a different engine: Ahrefs' Feb 2025 analysis of 1.4 million ChatGPT prompts found cited page titles had higher cosine similarity to the prompt (0.602) than uncited ones (0.484), and higher still when measured against fan-out style sub-queries (0.656). That study is about ChatGPT, not Copilot, so treat it as directional rather than as proof. The lesson survives the caveat: pages that name the question plainly hold up through query rewriting better than pages that are clever about it.

The grounding stage is the one people skip. Grounding checks, provenance checks and semantic similarity cross-checks run before the summary is composed, which means retrieved text has to hold up against the query and against the other retrieved passages. Content that cannot be corroborated tends to get dropped rather than repeated. If your central claim is a number that appears nowhere else, carries no date and cites no source, it is exactly the kind of passage that gets filtered out before the answer is written. Date your figures, attribute them, and make the corroboration easy to find.

What Microsoft actually tells you to do

The same documentation page publishes its indexing recommendations. There is nothing exotic in them, which is the point.

1. Keep sitemaps updated so Bing can discover URLs and content.
2. Use the IndexNow API, or Bing's URL or Content Submission API, to reflect website changes near-instantly instead of waiting for a recrawl.
3. Link every page from at least one other discoverable, crawlable page — Microsoft frames internal linking as a signal for determining site popularity.
4. Avoid unnecessary redirects.
5. Use dynamic rendering if the site is heavily client-side rendered.
6. Avoid nofollow and noindex tags on pages you want indexed.

Microsoft names three ranking parameters alongside that list: relevance to the intent behind the query, user engagement, and freshness. No proprietary AI signal, no markup requirement, no Copilot optimization layer.

Item three is where real sites fail. Orphan pages — live, published, linked from nothing — accumulate in any CMS where content gets created faster than navigation gets updated. They are also the default outcome of product listings reachable only through a JavaScript filter widget, and of campaign landing pages built outside the main template. Crawl your own site, sort by internal inlinks ascending, and look at everything sitting at zero. That list is usually longer than anyone expects.

Optimizing for generative AI search is optimizing for the search experience, and thus still SEO.

Google Search Central, Optimizing your website for generative AI features on Google Search, May 2026

That is Google describing its own surfaces, and it is self-serving. It also happens to describe Copilot accurately.

IndexNow, and how to switch it on

IndexNow is an open protocol for telling participating search engines that a URL has changed, rather than waiting for a crawler to notice. Microsoft's Bing and Yandex are the engines named as its co-developers, and Bing is the one that matters for this purpose.

The setup is small. Generate a key of 8 to 128 characters using only letters, numbers and dashes. Host it as a UTF-8 text file at your domain root, named after the key itself and containing the key and nothing else. Then notify the engine on publish — a GET for a single URL, or a POST carrying up to 10,000 URLs per request.

# 1. Key file at the domain root:
#    https://example.com/8fd2a1c47b3e4d59.txt
#    Contents: the key, nothing else.
8fd2a1c47b3e4d59

# 2. Single URL change
GET https://www.bing.com/indexnow?url=https://example.com/pricing&key=8fd2a1c47b3e4d59

# 3. Or batch, up to 10,000 URLs per POST
POST https://www.bing.com/indexnow
Content-Type: application/json

{
  "host": "example.com",
  "key": "8fd2a1c47b3e4d59",
  "urlList": [
    "https://example.com/pricing",
    "https://example.com/blog/new-post"
  ]
}

Most CMS platforms have a plugin that fires this automatically on publish; if yours does not, it is a few lines in the deploy pipeline. One honest caveat about the detail above: it comes from the IndexNow protocol's own documentation, while Microsoft's Copilot Studio page names the API without spelling out the mechanics. The protocol's origin story and the full list of participating engines are documented less consistently than the mechanics are, so do not promise a stakeholder that this pings every search engine on the web.

Fixing Bing fixes ChatGPT at the same time

This is why the work pays twice. ChatGPT's retrieval layer runs on search-provider results: OpenAI's publisher FAQ notes that ChatGPT search may still display a link and page title for a blocked site when the URL came from a third-party provider such as Bing. Ahrefs' Feb 2025 study found 88.46% of ChatGPT citations were derived from search results, well ahead of news, Reddit, YouTube or academic sources.

How tightly ChatGPT's citations track Bing's actual rankings is a weaker claim than most posts on this subject imply, so here is the honest version. One vendor study, from RivalHound, measured ChatGPT's cited results matching Bing's top organic results 26.7% of the time — real overlap, but nothing like a mirror, because ChatGPT re-ranks heavily after retrieval. The defensible conclusion is narrower and still worth acting on: being properly indexed in Bing is a shared prerequisite for both engines, even though strong Bing rankings do not buy you ChatGPT citations on their own.

Why bother, when Copilot's consumer share is shrinking

StatCounter, which measures outbound clicks from AI chatbots to websites, put Copilot at 3.19% of AI referral traffic in March 2026, down from a 5.18% peak in May 2025. As a consumer answer engine, Copilot is losing ground while ChatGPT holds 78.16% and Gemini climbs.

The consumer number is not the reason to care. Microsoft 365 Copilot is reported at 20 million paid enterprise seats as of Q3 FY26, with only about 4.4% of roughly 450 million eligible M365 seats attached. Those seat figures come from an aggregator compiling Microsoft's reporting rather than from a filing fetched directly, so treat the precision as indicative. The shape of the claim survives the hedge: Copilot's growth is happening inside the workday, in the same window as Outlook and Teams, on the desk of the person who signs your contract.

If you sell B2C, Copilot is a rounding error and you can deprioritize it honestly. If you sell B2B, Copilot is the answer engine embedded in your buyer's actual workflow, and the low attach rate says that installed base is early rather than finished.

The Bing indexing checklist you can finish this afternoon

1. Verify the domain in Bing Webmaster Tools; if you have Search Console, the import path takes minutes.
2. Submit the XML sitemap and confirm lastmod dates are real, not regenerated on every build.
3. Read the index coverage report and write down submitted versus indexed URLs. That gap is your actual Copilot problem.
4. Crawl the site, sort pages by internal inlinks ascending, and fix everything sitting at zero.
5. Search the templates for stray noindex and nofollow — staging defaults shipped to production are the classic cause.
6. Collapse redirect chains to a single hop.
7. If the site renders client-side, inspect what Bing actually sees and set up dynamic rendering if the rendered HTML comes back empty.
8. Turn on IndexNow, then publish something and watch how quickly it appears.
9. Confirm bingbot is not blocked at the CDN or WAF layer. Bot-management rules block it far more often than robots.txt does.

None of that is AI strategy. It is ordinary technical SEO applied to the index a decade of Google-first teams stopped maintaining, which is exactly why it is still available.