Bing Webmaster Tools had no way to show site owners when their pages got cited inside Microsoft Copilot until February 10, 2026, when Microsoft shipped an AI Performance dashboard in public preview. That gap mattered, because Copilot isn't one model answering from one index. It runs on a shifting mix of OpenAI models, Anthropic's Claude, and Microsoft's own in-house MAI models, grounded through Bing's search index rather than a separate AI crawl. Most AI visibility tools still track Copilot as a single, static assistant. The citation data Microsoft has actually published, and the mechanics behind it, say otherwise.
What actually powers a Copilot answer
Copilot combines three inputs, according to Microsoft's own support documentation: a user's Microsoft 365 work data through what Microsoft calls Work IQ, publicly available web content, and whatever files or links someone attaches to a prompt. The web leg is where GEO work matters, and it reaches Copilot through grounding rather than a Copilot-specific crawler. Grounding means Copilot retrieves current web content at answer time and cites it so a reader can check the claim.
That retrieval doesn't run on one model. Microsoft has started shifting some Copilot features, including transcription, onto its own MAI-Transcribe-1 model, and MAI-Image-2 is rolling out for image generation in Bing and PowerPoint, according to a July 2026 Bloomberg report. OpenAI and Anthropic's Claude still handle most production traffic inside Microsoft 365 Copilot. A brand that has already built a citation strategy for Claude is, without realizing it, partly optimizing for Copilot too.
Why Copilot's most-cited sources are retail giants, and why that doesn't shut out B2B content
Copilot's most-cited domains are dominated by retail and reference sites, not because niche B2B queries get ignored, but because shopping and lookup questions make up most of what people actually ask it. Ahrefs' Brand Radar analysis of more than 3 million Copilot queries in the United States found Amazon capturing a 16.8% mention share among the top 50 cited domains, with Walmart at 12.6% and Wikipedia at 7.6%. Health sites like Mayo Clinic and reference sites like Merriam-Webster round out the list, alongside a handful of low-authority sites, like scienceinsights.org, that rank surprisingly high despite modest backlink profiles.
None of that means a software company's content is invisible to Copilot. It means the measurement is about which specific queries a brand competes on, not overall citation share. Copilot's own search market share stays in the low single digits, as we've covered before, and a small share of search traffic doesn't translate to low stakes when a buyer's one relevant query lands on a competitor's page instead of yours.
What Bing Webmaster Tools' new AI Performance report actually shows
The AI Performance dashboard Microsoft shipped in February 2026 gives site owners four things they never had before: total citations, average cited pages, page-level citation activity, and a timeline of how citation patterns shift. None of it shows ranking or placement within an answer, only whether and how often a page got used. Microsoft's own description of the grounding-query metric is worth reading closely.
Grounding queries are the key phrases Copilot used when retrieving content it cited, and this metric is a sample of overall citation activity rather than a complete record.
That caveat matters for anyone treating the dashboard as a full picture. Microsoft has said it will keep refining the metric as it processes more data, so month-to-month swings in these numbers right now may reflect reporting changes as much as real shifts in what gets cited.
The Bingbot robots.txt rule that quietly affects your Copilot citations too
Because Copilot's web grounding rides on Bing's index, a robots.txt mistake that blocks Bingbot blocks Copilot's citations too, and one specific Bing crawler behavior catches sites off guard. Microsoft's Bing team has said that once a site gives Bingbot its own user-agent section in robots.txt, Bingbot ignores the sitewide default rules for every directive except Crawl-Delay. Any rule meant to apply to Bingbot has to be copied into its own section, or it simply won't be followed.
This is an old rule, documented by Bing's Frederic Dubut back in 2019, but it still catches sites that added rules for other AI crawlers more recently without checking whether a legacy Bingbot section already existed. The fix is a two-minute audit: search robots.txt for 'User-agent: bingbot' and confirm every directive meant to apply sitewide is repeated underneath it.
What Microsoft itself tells site owners to do about it
Microsoft's own content guidance in that same Bing blog post is specific enough to act on: use clear headings, use tables where the content is genuinely comparative, add FAQ sections, back claims with supporting evidence, update pages regularly, and keep entity details consistent across text, images, and video.
Clear headings and tables for genuinely comparative content
FAQ sections that answer questions directly
Supporting evidence attached to specific claims
Regular updates and consistent entity details across formats
Bing's team described the whole feature as an early step toward Generative Engine Optimization (GEO) tooling inside Bing Webmaster Tools, a notable admission from a company that spent two decades building Bing around classic search ranking. Microsoft also points to IndexNow as the mechanism that tells Bing when a page has actually changed, so a stale cached version doesn't keep getting cited after a correction. Structured data plays into several of these recommendations, particularly schema markup and FAQ schema, which earlier coverage breaks down block by block.
How this compares with ChatGPT's source selection
ChatGPT and Copilot both ground answers in live web retrieval, but they differ in one practical way that matters for anyone tracking citations across engines: Copilot now tells site owners what got cited, and ChatGPT doesn't offer an equivalent first-party dashboard yet. OpenAI's own source-selection criteria, covered in detail elsewhere, weight content structure and directness differently than Bing's grounding does.
That makes Copilot, for now, the only major assistant where a brand can log into a first-party tool and see its own citation count without relying on a third-party tracker to estimate it. Perplexity and Gemini publish even less about their own citation mechanics than Microsoft just did, which is reason enough to set up the Bing Webmaster Tools dashboard even for a brand that doesn't treat Copilot as a priority engine.
Frequently Asked Questions
What AI models does Microsoft Copilot actually use?
Copilot runs on a changing mix of models rather than one fixed engine. OpenAI's models and Anthropic's Claude handle most production traffic inside Microsoft 365 Copilot, while Microsoft has started shifting specific tasks, like transcription and some image generation, onto its own in-house MAI models where the economics work better. The exact mix varies by feature and changes as Microsoft rolls out more MAI models.
Does Microsoft Copilot use Bing search results to answer questions?
Yes. Copilot's web grounding runs through Bing's search index rather than a separate AI-specific crawl, which is why Microsoft's own guidance ties Copilot's citation behavior directly to robots.txt rules written for Bingbot. A page blocked from Bing's regular index is generally unavailable for Copilot's web-grounded answers too.
How can I check if my website is being cited in Copilot?
Microsoft added an AI Performance dashboard to Bing Webmaster Tools in February 2026, in public preview. It shows total citations, the average number of unique pages cited per day, page-level citation counts, and the grounding queries Copilot used to retrieve content, though Microsoft describes that last metric as a sample rather than a complete record.
What kind of content does Microsoft Copilot cite most often?
An Ahrefs analysis of more than 3 million Copilot queries found retail sites like Amazon and Walmart, reference sites like Wikipedia, and health sites like Mayo Clinic dominating the most-cited domains, reflecting the shopping and lookup questions that make up most Copilot usage. That skew doesn't mean niche B2B or technical content gets ignored; it means total citation share concentrates where query volume concentrates.
Does blocking Bingbot in robots.txt affect Copilot citations?
It can, in a specific way that trips up sites. Once a site creates a dedicated Bingbot section in robots.txt, Bingbot ignores the sitewide default rules for every directive except Crawl-Delay, so any rule meant to apply to Bingbot must be repeated inside its own section or it won't be followed.
Is Microsoft Copilot the same as Bing Chat?
Bing Chat was rebranded to Microsoft Copilot in November 2023, and the two names are no longer used interchangeably in Microsoft's current documentation. The web-grounding mechanics described in this guide, Bing's index and the AI Performance citation data, apply to Copilot across Microsoft's products, not a separate Bing-only chat experience.