How to Read Bing's AI Performance Report: Citations, Grounding Queries and Citation Share

Bing's AI Performance report in Bing Webmaster Tools shows how often Microsoft Copilot and Bing's AI summaries cite your pages, and which retrieval phrases triggered those citations. Open it for a verified site, export the three CSV files, and track citations, grounding queries and citation share by page. It measures Copilot only, and shows no clicks.

Microsoft launched the report in public preview in February 2026 and added four features in June. This guide walks B2B SaaS marketers through each metric, a step-by-step review routine, and the limits of the data. It is written for teams that want a measured view of AI visibility instead of screenshots of chatbot answers. Cekpr covers search and AI visibility for B2B software, and this report is the closest thing to first-party citation data that exists today.

What Is the AI Performance Report?

The AI Performance report is a dashboard in Bing Webmaster Tools that shows where AI answers cite your site. Microsoft's February 2026 launch post said it tracks citations in Microsoft Copilot, AI summaries in Bing, and select partner integrations. It shows which URLs are cited and how activity changes over time.

Microsoft Advertising's March 2026 post on the AI Performance dashboard lists the metrics. They are total citations, average cited pages, grounding queries and page-level activity. A grounding query is the phrase an AI system used to retrieve content before writing an answer. A newer view connects queries to pages, so one query can map to several pages and the reverse. Microsoft notes the data is a sample of overall citation activity, so treat the numbers as directional.

Any verified site in Bing Webmaster Tools can use it. If you have not verified your domain, do that first. You can import a Google Search Console property to speed up verification.

What Changed in June 2026?

Microsoft added four preview features on June 16, 2026. They are Intents, Topics, Citation Share and Compare. Microsoft's announcement of the new AI visibility insights describes each one.

Microsoft describes Citation Share as an observational metric. It says the metric is not a ranking system, does not reveal competitor domains, does not represent traffic share and does not score quality. Microsoft also warns that Intents and Topics labels can be broad in niche domains during preview.

How Do You Read the Report Step by Step?

Read it in a fixed order every month so you can compare results over time. Start with page-level data, then move to queries, then to the new classification views. This routine takes about thirty minutes.

Save every export. The tool has no API yet, and Microsoft has not given a timeline for one, according to analysis by Search Influence. A monthly folder of CSV files becomes your own history.

What Do Grounding Queries Tell You?

Grounding queries tell you how an AI system searched for content, which can differ sharply from what users typed. Search Influence reviewed more than 400 of them and found they looked machine-generated and keyword-dense. A single user question can produce several rephrased retrieval queries, a pattern called query fanout.

That has two consequences. First, raw citation counts can overstate real conversations. Search Influence estimated that one conversation may create three to five citation events, which is an estimate and not a Microsoft figure. Second, the queries work as a form of keyword research, because they show the language retrieval systems use for your topics.

Look for queries with commercial intent that cite a competitor's comparison page but not yours. Those are content gaps. Our guide on where AI answers get their sources for B2B software explains why brand-owned pages often lose to third-party ones.

What Does the Concentration of Citations Look Like?

Citations tend to cluster on a few pages. The Search Influence analysis of 20,000 Copilot citations covered 91 days and found 19,717 citations across 86 pages. One comparison page accounted for 69% of the total, and the top four pages produced 90%.

The same site saw a sharp decline. Citations averaged 1,520 per day in early December and fell to 34 per day by February, a 97% drop. That is one site's experience, not a benchmark, but it shows how quickly citations can fade.

The lesson is practical. A few deep, specific pages can carry your whole AI presence. Detailed comparison content with concrete specifics earned the citations in that case. Protect those pages with regular updates, and do not assume a citation spike will last.

What Are the Limits of This Data?

The limits are significant, and every report you share should state them. The data covers Copilot and Bing only, not ChatGPT, Perplexity, Gemini or Google AI Overviews. It shows citations but no click-through or traffic. It does not show which competitors were cited for the same queries.

It is also a preview. Microsoft says classification for Intents and Topics relies on evolving machine learning systems, and that citation patterns can shift with user behavior, model changes, freshness signals and partner refresh cycles. A change in Compare may therefore reflect factors beyond your content.

Do not treat Bing's numbers as a stand-in for all AI search. Pair them with Google's data. Our guide to measuring AI search visibility in Search Console shows the Google side, and Cekpr recommends reading both together.

How Should a SaaS Team Act on the Findings?

Turn the report into a short list of actions each month. Focus on pages and queries, not on vanity totals. The aim is to learn which content earns citations and why.

Structure alone will not guarantee citations. Our write-up on what Ahrefs' schema markup test means for SaaS teams shows how much the effect of one tactic can be overstated. Test changes one at a time and wait a full cycle before judging.

How Often Should You Check It?

Monthly is enough for most SaaS teams. Citation activity can swing day to day, and a monthly view smooths the noise. Check sooner after a major content release or a known platform change.

Set a calendar reminder and keep the same review window each time. Consistent windows make Compare far more useful. If you change the window, the comparison stops being fair.

Share a one-page summary with your team. Include the top five cited pages, the top ten grounding queries by commercial intent, and one recommended action. Short, regular reporting builds trust in the data without overselling it.

Build a simple scoring sheet to make reviews faster. List each cited page, its monthly citations, its top grounding query and the buyer stage that query suggests. Add a column for the last update date. Pages that earn citations but have not been updated in a year are your first refresh candidates.

Note which pages earn no citations at all. A pricing page or product page with zero citations may be perfectly healthy, because AI retrieval often favors explanatory and comparison content. Do not force every page to chase the metric.

Common Mistakes to Avoid

Teams make the same few errors with this report. They screenshot a spike and treat it as a trend. They read grounding queries as customer searches. They compare Bing's citation counts directly with Google impressions, which measure different things.

Another error is chasing the metric. A rise in Citation Share on a low-value query does not help revenue. Rank your queries by buyer value first, then watch share on the ones that matter.

Finally, do not report the numbers without context. Say which platform, which dates and that the data is a preview sample. That honesty keeps leadership from drawing the wrong conclusion.

Bottom Line

Bing's AI Performance report gives SaaS teams rare first-party data on AI citations, but only for Copilot and Bing, and without click data. Export monthly, study page concentration and grounding queries, and use Citation Share carefully. Treat the whole report as a directional signal.

Your next step is to verify your site in Bing Webmaster Tools, export your first CSV files and record a baseline. Review it again in 30 days and compare.