Bing AI Performance Report: How to Measure Copilot Citations
Use Bing Webmaster Tools' AI Performance preview to evaluate citations, cited pages and grounding-query evidence without mistaking citation for rank.
Bing Webmaster Tools’ AI Performance report shows how a verified site is cited across Microsoft Copilot, AI-generated Bing summaries and selected partner experiences. In public preview since February 2026, it reports total citations, average cited pages, sampled grounding queries, page-level citation activity and trends over time.
It is a source-visibility report, not a ranking report.
What the dashboard measures
Total citations
The total number of times site content is displayed as a source in supported AI answers during the selected period. It does not describe placement within an answer.
Average cited pages
The average number of unique pages cited per day. Bing warns that this aggregated measure does not indicate authority, ranking or a page’s role in one answer.
Grounding queries
Key phrases used during retrieval for answers that cited the site. Bing describes this as a sample, so treat it as directional topic evidence rather than a complete keyword report.
Page-level activity
Citation counts for individual URLs. This helps identify which assets are being reused and whether visibility is concentrated on a small number of pages.
Trend line
Citation activity over time across supported experiences. Annotate publishing, migrations and material content changes before interpreting movement.
Access and setup
Verify the site in Bing Webmaster Tools and open AI Performance. Confirm the selected property and date range. If no data appears, check Bing indexing, robots rules and the scale of eligible citation activity before concluding that reporting is broken.
Bing states that AI Performance respects content-owner preferences expressed through robots.txt and supported controls.
Turn the data into a backlog
Protect strong pages
For frequently cited URLs, review factual accuracy, evidence, freshness, canonical URLs and internal links. A high-citation page is a business dependency.
Expand grounding-query coverage
Cluster sampled grounding phrases by intent. Identify adjacent questions the page answers poorly, then improve the resource without stuffing phrases.
Find index-to-citation gaps
Compare important indexed pages with cited pages. A gap may reflect weak fit, thin evidence, poor clarity or insufficient authority. It is not proof of a penalty.
Diversify source assets
If one URL carries most citations, create complementary sources where the user intent is distinct: definitions, methods, comparisons, data and implementation guides.
Combine with other evidence
Pair the report with:
- Bing crawl and index data;
- IndexNow submission logs;
- page updates and server logs;
- prompt sampling in Copilot;
- GA4 referrals;
- conversion and branded-search trends.
The AI visibility dashboard guide shows how to place these layers in one reporting system.
Avoid false conclusions
Do not equate citation count with recommendation, sentiment or click value. Do not compare total citations across sites of very different size without normalization. Do not assume a grounding phrase is the user’s exact prompt. And do not claim IndexNow directly caused a Copilot citation.
A weekly workflow
- Export citations, pages and grounding-query samples.
- Compare the latest complete week with a four-week baseline.
- Flag material page or topic changes.
- Check affected URLs for technical or factual changes.
- create backlog items with owners and verification dates.
For a technical connection between discovery and AI experiences, see Bingbot, IndexNow and Copilot.
Frequently asked questions
Does AI Performance cover only Copilot?
No. Bing says the aggregated data covers Microsoft Copilot, AI-generated summaries in Bing and select partner integrations.
Is a grounding query the user’s prompt?
Not necessarily. Bing describes grounding queries as phrases used when retrieving cited content.
Does a citation count show where the source appeared?
No. Total citations count source display and do not indicate placement or role in a specific answer.
Is the report final?
It launched as a public preview, so fields and coverage may evolve. Record a last-reviewed date in internal documentation.
