Bing AI Performance: Free Citation Data and Grounding Queries

What does the AI Performance report show?

Microsoft's Bing team moved AI Performance in Bing Webmaster Tools from limited testing to public preview on 10 February 2026, making it available to every verified site at bing.com/webmasters/aiperformance, per coverage from Search Engine Journal and Search Engine Roundtable at the time. The report's original four metrics: total citations (how often your content appears as a source in AI-generated answers), average cited pages (the daily count of unique URLs referenced), page-level citation activity (which specific URLs get cited and how that changes over time), and grounding queries (the retrieval phrases behind those citations).

On 16 June 2026, Microsoft's own Bing blog announced four additions, in a post authored by four Bing/Microsoft AI product managers (Krishna Madhavan, Meenaz Merchant, Saral Nigam and Trishna Shah): Intents (classifying grounding queries into categories like informational, commercial, navigational, local and others), Topics (grouping related grounding queries into thematic clusters — the post's own example: individual queries about solar panels rolling up into a broader "Solar Energy" topic), Citation Share, and Compare. The same post, and a March 2026 update from Microsoft Advertising's blog, described a grounding-query-to-page mapping view: pick a query to see which pages were cited for it, or pick a page to see which queries drove its citations — a many-to-many relationship, since one query can map to several pages and vice versa.

To see this yourself: the report overview at bing.com/webmasters/aiperformance lays these out as separate panels for a verified site — total citations and average cited pages as headline numbers, a page-level citation table below them, and the grounding-queries list, Citation Share and Compare views as separate tabs alongside.

What is a grounding query?

Per Microsoft Advertising's own description, a grounding query is "the key phrases AI used to retrieve content that was cited, offering insight into how AI interprets user intent." It's generated by the system during retrieval, not a transcript of what a user actually typed — Search Engine Journal's coverage of the February launch paraphrased Microsoft's position as data that "represents a sample rather than complete citation activity." Treat a grounding query the way you'd treat a search term report with a confidence interval attached: directionally useful for understanding what topics your pages are being retrieved for, not a literal, complete log of every real user question.

How to map grounding queries to content gaps?

Three practical passes, once you have access to the report:

1. Scan grounding queries for topics you don't have a dedicated page for. A cluster of queries under a Topic you've never written a page targeting directly is a concrete content gap — the engine already knows the topic matters enough to retrieve for it. This is the same retrieval mechanic behind query fan-out: one buying question expands into several related retrieval queries, and grounding-query data is one of the few places you can see that expansion directly instead of inferring it.
2. Check which of your pages get cited for queries you didn't expect. This tells you how the engine is actually using a page, which sometimes differs from what you wrote it for — useful for deciding whether to expand that page or write a more targeted one.
3. Use Citation Share, not raw citation counts, to judge competitive position on a query. A query with 500 citations where you hold 5% share is a weaker position than a query with 50 citations where you hold 40% — raw totals alone invite the wrong comparison. Microsoft's own June 2026 post is explicit that Citation Share is "designed as an observational metric — not a ranking system," and that it doesn't expose competitor-level data directly, only your own share of the total.

What does it not cover?

Several real gaps, not quibbles:

• No click data. The February 2026 coverage from almcorp and Search Engine Journal both note the report shows citations, not whether anyone clicked through from one.
• No public API as of this writing. Exports exist as manual CSV/Excel downloads with active filters applied, per reporting from Search Engine Journal and a detailed how-to from Nick Blazer; Microsoft has described API access as being on its backlog, not shipped. The export control sits above the citation table — apply a date range or grounding-query filter first, since the download reflects whatever filter is currently active, not the full unfiltered dataset.
• Query-to-page detail doesn't survive export. At least one independent write-up (Nick Blazer) notes the export leaves out which specific query cited which specific page — the in-dashboard mapping view is richer than what currently comes out in a file.
• Topics labeling can be broad during preview. Microsoft's own June 2026 post acknowledges this directly, especially for specialized domains where a thematic cluster might lump together queries that a human expert would split apart.
• It's Microsoft's surfaces only — Copilot, Bing's AI summaries, and "select partner integrations," per the original February coverage. It says nothing about ChatGPT, Gemini, Claude, Perplexity or Google's AI Overviews and AI Mode.

How does it relate to Bingbot and IndexNow?

Citation requires crawling and indexing first — a page Bingbot has never fetched, or has fetched but not indexed, has nothing to be grounded from. IndexNow is Microsoft's own protocol for telling Bing a URL has changed without waiting for the next scheduled crawl; per Bing's own IndexNow documentation, it "ensures that search engines are aware of your website changes" and that "search engines will generally prioritize crawling these URLs versus other URLs they know" — though the same page is explicit that it "may take time for the change to reflect in search engines" and doesn't promise faster indexing, let alone faster citation.

Worth being precise here: Bing's own IndexNow documentation doesn't mention AI, grounding or citations anywhere, and that's still true as of this writing — other GEO commentary that draws the same IndexNow-to-AI-citation line (for example, write-ups recommending IndexNow "to help ensure updated content is discovered quickly by search and AI systems") does the same thing this post does: states it as the author's own inference, without a primary Microsoft statement backing it. So the connection to AI Performance is a reasonable, widely-repeated inference, not a stated fact — being crawled and indexed sooner is a precondition for being eligible to be cited sooner, but IndexNow itself only claims to speed up the crawl-priority signal, not any downstream AI-retrieval outcome. [VERIFY] if a future, more specific Microsoft statement ties IndexNow usage directly to citation or grounding-query performance; nothing opened for this post, or for several other independent write-ups checked alongside it, does.

FAQ

Is Bing's AI Performance report the same as a third-party AI visibility tool? No. It's first-party data from Microsoft about Microsoft's own surfaces (Copilot, Bing AI summaries), free, with no click data and no cross-engine view. A third-party visibility tool runs its own prompts across multiple engines but isn't first-party evidence of what Bing itself recorded. See First-Party AI Citation Reports vs Third-Party Visibility Tools for the full comparison.

Do I need a Bing Webmaster Tools account to see this data? Yes, and your site needs to be verified there, the same as any other Bing Webmaster Tools feature.

Is Citation Share the same as a ranking position? No. Microsoft describes it explicitly as an observational metric, not a ranking system — it shows your relative share of citations for a query, not a position 1-through-N the way a search ranking would.

Does showing up in grounding queries mean I'm cited in Copilot? Not automatically. A grounding query shows a retrieval phrase associated with your citation activity; the citation counts and page-level data are the more direct evidence that your content was actually used in an answer.

What to do next

1. Verify your site in Bing Webmaster Tools if it isn't already, and check whether AI Performance data is populated for your domain.
2. Export grounding queries and group them into topics you don't currently have a dedicated page for — that's your fastest content-gap list.
3. Track Citation Share, not raw citation counts, for your two or three most important category queries, and revisit it monthly using the Compare view added in June 2026.

ContextOwl sells one-time AI visibility reports. The AI Citation Snapshot is a free, single-question version of the same underlying question — whether a source is actually cited — for ChatGPT rather than Bing's surfaces.