Copilot Sources: How Microsoft Copilot Grounds Answers in Bing

Copilot answers from the Bing index. Microsoft documents Web Search and Grounding with Bing Search as generally available options on its agents API, and both are backed by Bing's index (Microsoft Learn, April 2026). The same index feeds Microsoft Copilot and the AI-generated summaries inside Bing.

That makes the chain simple to state:

• bingbot crawls your page.
• Bing decides whether to index it.
• At answer time, Copilot runs one or more grounding queries against that index.
• The model picks which retrieved pages to use and cite.

You influence the first two steps. Microsoft controls the last two. If you want the full picture of how this differs from Google, see which AI answers use which index.

One detail matters for measurement. Vercel's crawler study excluded Microsoft Copilot because it has no unique user agent. In practice, access for Copilot runs through bingbot, the same crawler that serves classic Bing search.

Microsoft documents more than most AI vendors about where sources come from. It documents much less about how they are chosen.

Documented

• The index. Copilot and Bing AI summaries ground on Bing, and Grounding with Bing Search is a GA developer product.
• The guidelines. Bing's Webmaster Guidelines describe what Bing indexes and what it treats as spam.
• The suppression controls. Bing added support for data-nosnippet in October 2025. Content with that attribute "is still indexed normally, but it will be excluded from snippets and AI summaries."
• The reporting. Bing Webmaster Tools' AI Performance report shows citations and grounding queries (more below).

Not documented

• Any ranking or selection criteria for which indexed pages Copilot cites.
• Any effect size for being in Bing on being cited by Copilot.
• How grounding queries are generated from a user's prompt.

Microsoft's own framing (May 2026) is that "search indexing was built to help humans decide what to read. Grounding indexing is being built to help AI systems decide what to say." It describes a purpose, not a formula. Treat anyone who claims to know Copilot's citation rules as guessing.

The link to ChatGPT is weaker still. OpenAI says ChatGPT search "sometimes partners with other search providers" and names Microsoft, but nobody has measured how much Bing presence moves ChatGPT results.

OtterlyAI published a 2026 experiment that isolates the index effect better than anything else in the public record. It launched two new domains with roughly 1,000 AI-generated posts each, then tracked Google, Bing and seven AI engines for about four months.

The results, from Otterly's write-up:

Microsoft's own AI Performance report credited capybara-world.com with 42,817 citations, more than ten times Otterly's prompt-based count. The time windows may not match, but it shows that a prompt tracker sees a sample.

Two lessons follow. An engine cannot cite what its index no longer holds, and Copilot kept citing what Bing kept. Being indexed is also not enough on its own: ChatGPT cited the sites effectively zero times, even while they ranked in Bing.

Keep the limits in view. It was one site per arm, on a low-competition topic (capybaras), run by a vendor that sells AI tracking. It shows the index mechanism working. It does not show that thin pages beat established sources in a commercial category, and mass-publishing cost both sites Google entirely. For the narrower question of pages missing from Bing, see whether Copilot can cite a site that isn't in Bing's index.

Everything you control sits at the access stage. These levers remove blocks. None of them makes a page more likely to be chosen.

Let bingbot in

Your robots.txt must not disallow bingbot. A blanket AI block written for training crawlers can catch it by accident. A minimal check looks like this:

User-agent: bingbot
Allow: /

Make sure the network agrees with robots.txt

A firewall or CDN can return a 403 to a crawler that robots.txt allows. The operators document this mechanism, though nobody has measured how often it happens. The AI Crawler Checker requests a page as each crawler and shows which ones your firewall turns away.

Check for accidental suppression

• noindex removes the page from Bing's index, and so from Copilot.
• nosnippet and max-snippet:0 limit what can be shown.
• data-nosnippet keeps a block out of Bing's snippets and AI summaries while it stays "available for ranking."

These often arrive inherited from a template. Search your HTML for them.

IndexNow, scoped honestly

Bing consumes IndexNow, and Microsoft has asserted, without a published study, that it helps AI systems reference the current version of a page. The IndexNow specification never mentions AI, and a 200 response confirms only that the ping was received. Treat it as free Bing hygiene with no citation promise.

The broader playbook is in getting into Microsoft's AI answers.

Bing Webmaster Tools' AI Performance report has been in public preview since February 2026 and was expanded in June 2026. It covers Microsoft Copilot, AI-generated summaries in Bing and select partner integrations. It shows Total Citations, Average Cited Pages, page-level citation activity, Citation Share, intents, topics and grounding queries.

Grounding queries are the unusual part. They are the phrases the AI actually ran against the index when it retrieved content it then cited. No other vendor exposes the real retrieval queries. Every prompt-based tracker tests prompts it chose, which are a proxy for queries nobody else can see. Our playbook on exporting and analyzing Bing grounding queries walks through using them.

Carry three caveats:

• Microsoft says the grounding query data is a sample of citation activity. The note sits in that subsection, so do not assume the whole report is sampled. More in is Bing's AI Performance data complete or sampled.
• Microsoft calls Citation Share "an observational metric, not a ranking system or a competitive scoreboard."
• As of the evidence date (August 2026), the report had no API. Microsoft has indicated API access during 2026, so check before you build a workflow around exports.

Work through the gates in order, because each one depends on the last.

1. Verify the site in Bing Webmaster Tools. Confirm your key pages are indexed. Google Search Console tells you nothing about Bing.
2. Test access. Check that robots.txt allows bingbot and that your CDN returns a 200, not a 403 or a challenge page.
3. Remove accidental suppression. Audit noindex, nosnippet, max-snippet and data-nosnippet on pages that carry your pricing and product facts.
4. State your facts plainly in the HTML. A specific price or one-line definition on a fetchable page gives Copilot a sentence it can quote. If Copilot already states something wrong about you, see fixing wrong facts in Copilot answers.
5. Read the AI Performance report monthly. Compare grounding queries with the questions you assumed buyers ask.

What stays out of your hands, as of October 2026: which indexed pages Copilot selects, how it rewrites prompts into grounding queries, and whether any given answer cites you. Fix the gates, then measure.