Bing AI Performance provides first-party figures: Microsoft itself counts a website’s AI citations in the AI answers people were actually shown, not in selected test questions. The figures appear in Bing Webmaster Tools, Microsoft’s tool for website owners. At launch, Microsoft wrote that site owners could see for the first time how often their content is cited in generative answers, and called the report an early step toward tooling for Generative Engine Optimization (GEO).

The report covers citations in Microsoft Copilot, in AI-generated summaries in Bing, and in select partner AI integrations that Microsoft does not name. A citation is counted when content is visibly given as a source. As in Bing itself, only content that Bing may include in its index can appear, and Microsoft states that Bing respects site owners’ preferences set in robots.txt and other supported controls.

To see the report, you need a Bing Webmaster Tools account and a verified website, i.e., proof that you own the site. A site already verified in Google Search Console, a free Google service for website owners, can be imported from there. According to Microsoft, data usually appears about 48 hours after a site is added.

What the report shows

For the selected period, the report shows:

  • Total Citations: how often the site’s content was given as a source. This count is one form of citation rate.
  • Cited Pages and Average Cited Pages: how many different pages of the site were cited on a given day, and how many per day on average.
  • Pages: which individual URLs were cited and how often—in other words, which content actually serves as a source.
  • Grounding queries: short phrases the AI used to retrieve the content it later cited.
  • Timeline: how citation volume changes over time.

Microsoft refreshes the data daily with a short processing delay. Periods of 7 days, 30 days, or 3 months are available, as well as custom ranges within the available data; according to Microsoft, historical data is available only for these ranges. Grounding queries, pages, and the timeline can be exported as CSV or Excel files. To follow trends over longer periods, it therefore makes sense to export the figures regularly, for example once a month: total citations, the most-cited pages, and the most important grounding queries.

On June 16, 2026, Microsoft began rolling out four additional analyses in preview. Intents classify grounding queries by the purpose behind them, such as informational, commercial, local, or comparison. Topics group related grounding queries into broader themes, such as solar energy. Citation Share shows a site’s percentage of all citations shown for a grounding query—a form of AI share of voice, though without revealing which sites hold the rest. Compare overlays an earlier period on the current chart. Microsoft assigns intents and topics automatically with AI, and according to Microsoft, the labels do not always fit, especially for ambiguous queries and highly specialized fields.

Grounding queries: how the AI retrieves content

When an AI system bases an answer on content it retrieves from the web at the moment of the question, this is called grounding. The grounding queries in the report are the short phrases the AI used to retrieve content from the site that it then cited. Microsoft groups similar ones into generalized expressions. Its example: if a site is cited in answers about solar energy, that activity may appear under grounding queries such as “solar energy” or “solar panels.”

Grounding queries are neither questions as people type them nor complete prompts. An AI system can split a single question into several subqueries and retrieve content for each, which is known as query fan-out; the report does not show such subqueries individually, only in this grouped form. One grounding query can relate to several pages, and one page to several grounding queries. In March 2026, Microsoft linked the two in the report: selecting a grounding query shows the pages cited for it, and selecting a page shows its grounding queries. Microsoft also notes that the individual AI services and partners may determine grounding queries differently. Still, they are useful hints: they show the topics for which a website already serves as a source and can suggest which topics belong in a prompt set for your own measurement.

What the report does not show

According to Microsoft, a citation only means that content was visibly given as a source. The report shows no clicks, visits, or other engagement; how many people reach the website through links in AI answers is captured in web analytics as AI referral traffic. Nor does the report say anything about a page’s role in an answer: it counts how often pages are cited, not how important they were to an answer or where they appeared.

The figures are based on a sample and are not a complete log of every citation; Microsoft may refine results as more data is processed. The report is therefore designed for spotting trends over time rather than counting individual answers precisely. Even a trend cannot be attributed to a single cause, according to Microsoft: citation volume can change because people ask different questions, because content changes, or because the AI systems and models involved are updated.

Finally, the report covers only Microsoft’s AI services and its partners. For Google’s AI Overviews and AI Mode, Google Search Console offers a counterpart, the generative AI performance report. It counts impressions: how often links to the site were shown in those AI answers. How other AI systems portray a brand, and whether they mention it without a link, only becomes visible through GEO monitoring with a fixed set of your own questions.