Bing Webmaster Tools launches AI Performance in public preview — a citation dashboard, not a ranking signal

What the public preview is and what its metrics actually mean

The Bing Webmaster Tools team announced the public preview of AI Performance on February 10 2026, describing it as a new set of insights that shows how publisher content appears across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations, with visibility into which URLs are referenced and how citation activity changes over time (C1). The dashboard is positioned by Microsoft as an early step toward Generative Engine Optimization (GEO) tooling in Bing Webmaster Tools, and the team states they will continue working with publishers and the webmaster community to improve inclusion, attribution, and visibility across both search results and AI experiences, which means metric definitions and supported surfaces may change (C8).

Four metrics are named in the post. Total Citations shows the total number of citations that are displayed as sources in AI-generated answers during the selected time frame, and the post is explicit that this highlights how often content is referenced by AI systems, without indicating placement or presentation within a specific answer (C2). Average Cited Pages shows the average number of unique pages from your site that are displayed as sources in AI-generated answers per day over the selected time range, and the post states that this does not indicate ranking, authority, or the role of any page within an individual answer (C3). Grounding queries is presented as a sample of overall citation activity that the team says they will continue to refine as additional data is processed, so individual grounding-query counts are not a full inventory (C6). Page-level citation activity shows citation counts for specific URLs over a date range, and the post is explicit that this reflects how often pages are cited, not page importance, ranking, or placement (C7). The Important Note in the announcement states that Bing respects all content owner preferences expressed through robots.txt and other supported control mechanisms, so publishers' existing crawl controls continue to govern AI surface inclusion alongside this new reporting (C4). For publishers already working in SEO & Webmaster tools, the practical effect is a new reporting surface, not a new permissioning surface.

What changed, what did not, and what the post does not authorize

The change is reporting-only. The post introduces a consolidated view of when a site is cited in AI answers and offers a timeline that shows how citation activity changes over time across supported AI experiences, but the language throughout the post is careful to separate citation counts from ranking, authority, and placement. The dashboard measures reference frequency, not position within an answer; it does not declare a page important because it is cited, and it does not declare a site authoritative because it is cited more often than another. Page-level citation activity reflects how often pages are cited, not page importance, ranking, or placement (C7), and Total Citations is defined without indicating placement or presentation within a specific answer (C2). Average Cited Pages is defined as not indicating ranking, authority, or the role of any page within an individual answer (C3). The post does not announce any new ranking factor, any change to retrieval, any change to grounding models, or any change to how Microsoft Copilot or Bing AI-generated summaries decide which sources to cite. It does not announce a rollout window beyond the public preview itself, and it does not authorize treating citation counts as a Search Console–style performance metric for AI surfaces.

The controls story is also unchanged. Because Bing respects all content owner preferences expressed through robots.txt and other supported control mechanisms, publishers' existing crawl controls continue to govern AI surface inclusion alongside this new reporting (C4). That is consistent with the post's "Evolving AI Performance with the Webmaster Community" framing, in which the team frames the release as an early step toward Generative Engine Optimization (GEO) tooling in Bing Webmaster Tools and commits to continued work with publishers and the webmaster community, with metric definitions and supported surfaces still subject to change (C8).

Real alternative explanations, confounders, and what the numbers can hide

Several alternative explanations are worth naming before any publisher reacts. First, citation counts are a sample, not a census: Grounding queries is shown as a sample of overall citation activity and the team states they will continue to refine this metric as additional data is processed, so individual grounding-query counts are not a full inventory (C6). A page that "drops out" of grounding queries may still be cited; it may simply not surface in the sampled phrase set.

Second, citation count is not position. The post defines Total Citations without indicating placement or presentation within a specific answer (C2) and defines Average Cited Pages without indicating ranking, authority, or the role of any page within an individual answer (C3). A heavily cited page can sit deep inside an answer with weak prominence, and a lightly cited page can be the lead source.

Third, coverage is partial by design. The dashboard reports Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations (C1); citations from third-party AI products that are not in the supported set are not represented. Fourth, the metrics are themselves labeled an early step toward Generative Engine Optimization (GEO) tooling, with the team stating they will continue working with publishers and the webmaster community and that metric definitions and supported surfaces may change (C8).

Comparisons across time windows that cross a definition change will be confounded. txt and other supported control mechanisms continue to govern whether a URL is eligible to be cited at all (C4), so a low citation count on a heavily disallowed URL is an artifact of policy, not a signal of low quality. Page-level citation activity, finally, reflects how often pages are cited and is explicitly not a statement of page importance, ranking, or placement (C7), which means using the per-URL list as a "winners" leaderboard will misread what the post actually says.

Knowledge Delta: mechanism, decision, and a falsifiable follow-up signal

Mechanism, as stated in the post, is straightforward: the dashboard aggregates already-recorded citation events from supported AI surfaces and presents them as four metrics, with the explicit caveat that it does not measure ranking, authority, or placement within an answer. Total Citations aggregates citations across the selected time frame (C2); Average Cited Pages averages unique cited URLs per day across the selected time range (C3); Grounding queries exposes a sample of the key phrases the AI used when retrieving content that was referenced (C6); page-level activity exposes the per-URL count over the date range (C7); and all of this sits on top of existing robots.txt and supported control mechanisms (C4). The release is framed as an early step toward GEO tooling, with the team committing to continued iteration on inclusion, attribution, and visibility, which means metric definitions and supported surfaces may change (C8).

Decision: do not treat AI Performance as a ranking dashboard, a quality dashboard, or an authority dashboard. Use it as a citation-frequency dashboard. It tells you which URLs of yours are being referenced and how often, on a sampled basis; it does not tell you how prominently, how authoritatively, or how well. Falsifiable follow-up signal: if a future revision of the post or the product removes the "not ranking, not authority, not placement" caveats from the definitions, that would convert this from a citation-frequency signal into a quality or ranking signal and would require a full re-read of any conclusion built on this preview. Until then, the supported surfaces and the metric definitions remain the most consequential unknowns (C8), and the grounding-query sample is a known approximation (C6).

Public Action Brief: action level, what to do, what not to change, and what to measure

LABEL: ACTION LEVEL — Watch only LABEL: HIGH IMPACT CHANGE — NO LABEL: WHAT TO DO NOW — Open the AI Performance dashboard and record a baseline of Total Citations, Average Cited Pages, a Grounding queries sample, and the page-level list for the current date range; cross-check that URLs you expect to be cited (for example, your product, pricing, and how-to pages) appear in the page-level list; confirm that your robots.txt and supported control mechanisms still reflect the AI surface inclusion you intend (C1, C2, C3, C4, C6, C7). LABEL: WHAT NOT TO CHANGE YET — Do not re-prioritize your content roadmap based on citation counts; do not de-publish or rewrite pages because they are under-cited; do not add new structured data purely to chase grounding-query phrases, since the sample is partial and metric definitions may change (C6, C7, C8); do not loosen robots.txt or other supported control mechanisms to chase more citations, since those controls still govern inclusion (C4). LABEL: MEASUREMENT BASELINE — The current pre-action snapshot from the four metrics, captured as Total Citations, Average Cited Pages per day, a saved Grounding queries sample, and a saved page-level citation activity list, against the date range you select on first open (C2, C3, C6, C7). LABEL: MEASUREMENT METRICS — Total Citations over the selected time frame (C2); Average Cited Pages per day over the selected time range (C3); Grounding queries sample as published, with the explicit caveat that it is a sample and not a full inventory (C6); page-level citation counts for your URLs over the date range, treated as frequency, not importance (C7). LABEL: MEASUREMENT SEGMENTS — Site-wide totals, the per-URL page-level list, and the Grounding queries sample; segment internally by content type (for example, product, support, editorial) without ranking those segments against each other on the basis of citation count. LABEL: OBSERVATION WINDOW — The single public preview announcement and the metric definitions published with it on February 10 2026; subsequent iterations are not yet documented in the frozen sources. LABEL: WHAT WOULD CHANGE THIS CONCLUSION — A documented change to the metric definitions that removes the "not ranking, not authority, not placement" caveats, a change to the supported AI surfaces, or a change to how Grounding queries is sampled (C6, C7, C8); a documented change to how robots.txt and other supported control mechanisms are honored on AI surfaces would also change the conclusion (C4). LABEL: WHEN TO REVIEW — When the Bing Webmaster Tools team publishes an updated definition for any of the four metrics, adds or removes a supported AI surface, or changes the Grounding queries sampling method, whichever comes first (C8). LABEL: APPLICABILITY — Publishers of any size with verified Bing Webmaster Tools access whose content is eligible to be cited by Microsoft Copilot, AI-generated summaries in Bing, or the named partner integrations (C1, C4). LABEL: RISK BOUNDARY — The post does not authorize treating citation counts as a quality, ranking, or authority signal; it does not announce a ranking change, a retrieval change, or a new crawl directive; existing robots.txt and supported control mechanisms continue to govern inclusion (C2, C3, C4, C7).

ACTION LEVEL: Watch only WHAT TO DO NOW: Monitor the affected cohort against the frozen Search Central evidence before changing production systems. WHAT NOT TO CHANGE YET: Do not rewrite templates, canonicals, or ranking-oriented site structure solely because of this observation. WHAT WOULD CHANGE THIS CONCLUSION: An official confirmed Google Search Central change notice for this subject. WHEN TO REVIEW: 2026-08-25