By meeting the eligibility conditions Microsoft has written down, because unlike every other major AI assistant, Copilot has published rules. The Bing Webmaster Guidelines now cover Bing search, Copilot and grounding API results in one document, and they state plainly which practices make a URL eligible for grounding results and citations. Most of the work is technical and unglamorous, and two of the most common ways to lock yourself out of Copilot are a single meta tag.
Why does Copilot deserve separate attention from ChatGPT?
Because it runs on an index you can see into and influence directly, and because Microsoft tells you the rules. Everything we have written about getting cited by ChatGPT, appearing in Perplexity, getting cited in Google Gemini and getting cited in Grok is inference from observed behavior. With Copilot, you are reading published guidance from the company that operates the retrieval layer.
Microsoft also gave this layer a name. In a post published on 12 February 2026, Elevating the Role of Grounding on the AI Web, Bing defined grounding as the system connecting AI models to current, authoritative information so answers draw on more than training data, and introduced generative engine optimization as the practice of understanding how content takes part in AI-driven experiences. The guidelines are blunt about the ceiling: SEO does not guarantee rankings, and GEO does not guarantee grounding or citations. Treat anyone who promises either as selling something they cannot deliver.
What are the two tags that quietly block you from Copilot?
noarchive and nocache, and most teams have no idea they are there.
Per Microsoft's own guidelines, noarchive prevents content from being used in Copilot responses and grounding results at all. nocache is softer and still damaging: it limits Copilot to the URL, title and snippet, which reduces how deeply your page can be cited. nosnippet and data-nosnippet stop Bing from displaying captions and may limit citation quality, and the guidelines point to a data-snippet attribute if you want to control which text is quotable rather than suppress all of it. noindex removes a URL from Bing search, Copilot experiences and grounding results together.
This matters because these tags often arrive by accident. A security review adds noarchive sitewide. A staging template ships to production. A plugin sets nocache on article pages. Nobody connects it to AI visibility, because until Microsoft wrote it down, nobody could. Checking your page source for those four directives is a ten-minute job and it is the single highest-value action in this article.
What does Microsoft say actually makes a page citable?
Content that stands on its own. That phrase, in various forms, is the thread running through the guidelines, and it is more specific than it sounds.
The guidelines ask for facts and definitions stated explicitly, key statements that do not depend on implied context, and important information visible on the URL itself rather than assembled from elsewhere. They ask for each URL to focus on a single primary topic and to avoid mixing unrelated concepts. They ask for essential information near the top of the page, with long introductions avoided. They ask for consistent, unambiguous naming of people, organizations, products and locations.
Read as a checklist for a company page, that means:
- State the claim, do not gesture at it. A sentence a retrieval system can lift and verify beats three paragraphs of context. This is the same discipline that makes a page quotable to a journalist.
- One question per page. A page that answers one thing well is more likely to be selected than a page that covers five things adequately. Our own article structure follows this on purpose.
- Answer in the first paragraph. Then explain. The reverse order, context first and answer last, is how most corporate pages are written and it is working against you.
- Name yourself the same way every time. Inconsistent company, product and person names split the entity, and the guidelines are explicit that ambiguous references hurt citation accuracy.
The guidelines also say images and video should reinforce the text rather than be the only place the information lives, with alt text, captions, transcripts or structured data to support understanding. A key number that exists only inside a graphic is a number Copilot cannot cite.
Which technical basics does Microsoft tie directly to AI eligibility?
The ordinary ones, which is the point. Microsoft's position is that the crawling and indexing foundation under traditional search is the same foundation under Copilot, so crawl efficiency, canonical URL consolidation, clear structure and trust signals support grounding eligibility as well as rankings.
Specifically, the guidelines ask you to:
- Make URLs discoverable through IndexNow submissions, XML sitemaps, crawlable internal links and external links from relevant sites.
- Use IndexNow when URLs are added, updated or removed, and prefer streaming submissions over batches, because timely notification reduces outdated URLs being referenced in Copilot responses.
- Keep sitemaps to canonical URLs only, remove deleted and redirected URLs promptly, and include accurate
lastmodvalues. - Consolidate duplicates, because duplication reduces Bing's confidence when choosing which URL to cite.
- Use 301 redirects for permanent moves, and reserve 302s for changes shorter than two days.
- Let Bingbot render the page, rather than hiding the substance behind client-side rendering.
That last point is worth sitting with. If your content only exists after JavaScript runs, you are relying on a renderer to find it. The same logic applies to an llms.txt file, which is a voluntary convention rather than a requirement: useful as a signal, no substitute for the content being in the HTML.
How do you measure whether any of this worked?
Through Bing Webmaster Tools, which now reports AI citations directly. Microsoft described the AI Performance dashboard in a post on 23 March 2026: it reports total citations, the grounding queries AI used to retrieve the cited content, which specific URLs are cited most often, and how that activity changes over time.
Grounding queries are the useful part. They tell you the phrasing a retrieval system used to reach your page, which is often not the phrasing you optimized for, and that gap is the most actionable thing in the report. Pair it with Google's own AI performance report in Search Console and you have the two free, first-party views of AI visibility that currently exist. Neither covers ChatGPT, Claude or Perplexity, which is why third-party AI visibility tools still have a job.
The guidelines add one framing worth repeating to anyone judging your performance on traffic alone: content can surface across Bing experiences without a click, so falling clicks do not always mean falling visibility, and impressions, indexing status and citation eligibility are the things to watch alongside sessions.
What will get you excluded?
The abuse list in the guidelines includes the familiar ones, cloaking, link schemes, keyword stuffing, scraped or lightly rewritten content, thin affiliate pages and misleading structured data. Two entries are newer and specific to this moment.
The first is content generated at scale without oversight or editorial review, which Microsoft says may be excluded from indexing. The second is prompt injection and AI manipulation: text added to a page to interfere with the models Bing and Copilot use may result in reduced visibility or removal. If anyone has suggested hiding instructions in your markup to make an AI assistant recommend you, that is now a named violation with a stated consequence.
None of this is a reason to be cautious about publishing. It is a reason to publish things a person would want to read. The pages that earn citations are the ones that say something checkable, and the fastest way to have something checkable to say is earned coverage and original data rather than more pages about yourself.
The direct answer, for anyone who read this far
Check your HTML for noarchive, nocache, nosnippet and noindex first, because those tags block or degrade Copilot citations outright. Then make each important page answer one question explicitly, near the top, with named entities and facts a system can verify without leaving the page. Keep IndexNow, sitemaps and canonicals clean, and watch grounding queries in Bing Webmaster Tools to see what Copilot thinks you are about. If you want that audit done properly, along with the coverage that gives you something worth citing, get in touch.
