Key takeaways
- Citation counts describe observable use in Microsoft AI experiences; they are not an authority score, ranking, or click total.
- Grounding-query themes can reveal information gaps without justifying a separate page for every wording variation.
- Group the data by cited URL, intent, topic, and commercial role before deciding what to improve.
- Connect AI-search diagnostics with analytics and CRM outcomes while keeping attribution limits visible.
01
Know what the report can—and cannot—tell you
Bing Webmaster Tools’ AI Performance report shows when pages from a verified site are cited in Microsoft Copilot and Bing’s AI-powered experiences. It can surface cited pages, grounding queries, and intent categories. That is valuable because generated discovery often hides the long chain between a person’s wording and the sources used to compose an answer.
Bing is equally clear about the boundary: citation count is not a ranking, an authority score, or a direct count of referrals. Start the dashboard with that note. Otherwise a rising line can become an unearned business claim before anyone asks what changed downstream.
02
Turn grounding queries into editorial decisions
Export or record query themes, then cluster them by the job a person is trying to complete: learn, compare, solve, research, create, navigate, buy, or find something local. Look for a missing explanation, an outdated fact, a weak example, or a useful comparison the cited page does not yet handle well.
Do not turn each phrase into a URL. Ten similar queries may call for one clearer section, a table, or a linked supporting guide. A new page earns its place when it helps with a distinct decision and contains evidence or an artifact the current page cannot accommodate cleanly.
- Strengthen an existing passage
- Add a dated example or source
- Create a genuine comparison
- Publish an implementation artifact
- Leave the topic alone when the current answer is already sufficient
03
Compare cited and uncited pages carefully
Review structure, freshness, specificity, source quality, internal links, entities, and unique evidence. Avoid declaring a single factor causal. Citation selection can differ across engines and interfaces, and the same engine may choose differently when query phrasing, language, or current information changes.
A useful monthly review pairs examples with aggregates. Read several actual passages from strongly cited pages and several strategically important pages with little visibility. The qualitative review often reveals a missing definition or buried answer that a citation total cannot explain.
04
Build a two-layer scorecard
The discovery layer contains cited URLs, citation trend, query themes, intent mix, representative answer checks, crawl health, and freshness. The commercial layer contains qualified AI referrals where identifiable, assisted conversions, branded search change, pipeline touchpoints, and sales feedback. Keep the layers connected but do not pretend they are the same measurement.
Annotate major content releases and technical changes. Review a stable question panel on a fixed cadence, then add exploratory questions separately. This makes the signal more comparable over time without suggesting that an inherently variable generative system has become a laboratory instrument.
Swipe to compare every column
| Signal | What it helps diagnose | What it does not prove |
|---|---|---|
| Citation count | Whether pages are being used as sources | Ranking, authority, traffic, or revenue |
| Grounding queries | Themes and intents behind source selection | Exact public prompt volume |
| Cited pages | Which URLs supply useful passages | That the whole page is high quality |
| AI referrals | Some identifiable visits from AI experiences | The complete influence of generated discovery |
Primary sources and further reading
Use the source material to validate details against your own context and current platform configuration.
- Bing Webmaster Tools: AI Performance
- Bing Webmaster Guidelines
- Empirical analysis of generative search engines
This field note follows the XenGrowth editorial policy: primary sources where available, visible limitations, material review dates, and no invented first-hand experience.
Stay with the problem
