I am a fan of Bing Webmaster Tools. I’ve said it multiple times, and I’ll say it again: I think they’ve come a long way in the last few years and have really improved their product.
Today, we’re looking at a relatively recent feature: the AI Performance Report. We’ll use one of our sites to show you real (anonymized) data and walk you through it. Let’s go!
What is Bing Webmaster Tools AI Performance?
AI Performance is a free report inside Bing Webmaster Tools that shows when pages from a verified website are cited in supported AI-generated answers. Microsoft introduced the public preview in February 2026. The first version reported total citations, average cited pages, grounding queries, page-level citation activity and changes over time.
In June, Microsoft expanded the report with four additional capabilities:
- Intents, which classify grounding queries using labels such as Informational, Commercial, Research, Comparison, Planning and Local.
- Topics, which group related grounding queries into broader themes.
- Citation Share, which estimates the percentage of citations attributed to your site for a specific grounding query.
- Compare, which overlays two periods so that changes in citation activity are easier to see.
According to Microsoft’s AI Performance documentation, the report covers Microsoft Copilot, AI-generated summaries in Bing and “select partner AI integrations.” We will return to that deliberately vague third category later.
Here are the most important metrics covered
| Metric | What it means in normal language | How we would use it |
| Total Citations | The number of times content from the site was visibly referenced as a source | Watch broad direction, not business impact |
| Average Cited Pages | The average number of unique site pages cited per day | See whether citation visibility is concentrated or spreading |
| Grounding Queries | Grouped phrases used when retrieving cited content | Understand how the system associates pages with subjects and intents |
| Page-level citations | Citation counts attached to individual URLs | Find unexpected winners and underperforming priority pages |
| Citation Share | The site’s percentage of citations for a particular grounding query | Track relative presence for queries that genuinely matter to the business |
The word grounding makes this sound more complicated than it needs to be. An AI system needs sources on which to base an answer. The retrieval phrases used to find those sources are grounding queries. They are not necessarily the words typed by a person.
7 things we love about Bing AI Performance
1. It gives publishers first-party AI citation data for free
Let us start with the simplest thing Bing gets right: verified site owners can see first-party AI citation data without buying another tool. At a time when AI visibility products are multiplying almost as quickly as their dashboards, free access is more important than ever.
The first number most people will notice is Total Citations. It is large, clean and easy to screenshot. It is also extremely easy to misuse. A citation means that a page was visibly referenced or shown as a source in a supported AI answer. It does not mean that the citation was prominent, that the user opened it or even that the page played a decisive role in the answer.
Microsoft is unusually direct about this. It says AI Performance does not measure ranking, authority, importance or performance. The report counts citations.
That makes Total Citations useful as a baseline and a directional trend. It gives a team a sensible place to start asking questions, even if it is not a substitute for traffic, readership or commercial value.
2. The page-to-query mapping is genuinely useful
The Grounding Queries view is where AI Performance becomes more than a decorative chart.
Each row shows a short phrase associated with retrieval and citation activity. You can select a query to see the pages cited for it, or select a page to see its associated grounding queries. A query can map to several pages, and one page can map to several queries.
For a publisher, this creates three useful investigations.
3. Grounding queries show how Bing understands your content
Grounding queries give publishers something most AI visibility reports struggle to provide: a view of the retrieval context behind a citation. They show the subjects and tasks Microsoft’s systems associate with a page, which can expose a useful angle the editorial team had not considered.
There is an important boundary. Grounding queries are not full user prompts. Microsoft describes them as grouped, generalized phrases representing activity across multiple AI answers. They may look short or vague because the dashboard is summarizing a larger set of interactions.
Anyone presenting these phrases as “the prompts people used” is claiming more than the data provides.
That still makes the phrases useful for understanding broad retrieval context, but unsuitable for claims about exact audience language. We would not copy them into a content brief as if they came from a conventional keyword report. Their value is in the relationship they reveal between a subject and a cited page.
4. Intents and Topics make a messy dataset easier to understand
The Intents and Topics additions make a large export easier to scan.
Intents classify grounding queries into categories extending beyond the traditional informational, navigational and transactional model. Bing includes categories such as Learn and Solve, Research, Comparison, Planning, Utility, Creation and Conversational.
That is useful because the same subject can support very different tasks. Someone learning what a product does is not at the same stage as someone comparing it with two alternatives.
Topics work one level higher. They group related grounding queries into broader thematic clusters so that a publisher can see whether citation activity is building around a subject rather than a single phrase.
We would use both as filters for an editorial review. We would not allow either classifier to decide the editorial plan automatically.
Microsoft says the labels are generated by evolving AI and machine-learning systems. Ambiguous queries can be assigned an imperfect intent, while specialist queries can land in topics that feel too broad. That is an honest limitation and one we expect to encounter frequently on technical publications.
5. Citation Share adds context that raw totals cannot
Citation Share is the percentage of citations attributed to your site out of all citations shown for the same grounding query.
This is more useful than a raw total because it introduces context. A site can gain citations while losing share if the overall citation space is expanding faster. It can also have a modest total but a strong share within a narrow, valuable query. That is a meaningful improvement over celebrating a citation count in isolation.
Our preferred use would be to select a small set of grounding queries that matter to the publication and monitor those consistently. Looking for the highest percentage anywhere on the site will mostly produce vanity findings.
6. Compare gives content updates a proper before-and-after view
Compare overlays a previous period as a dashed line against the current period. You can compare standard periods or choose custom ranges.
This makes the report far easier to use after updating content. Record the publication or update date, allow time for recrawling and processing, then compare equivalent windows before and after the change.
Microsoft lists several other reasons citation patterns can move: changes in user demand, model updates, partner refresh cycles, freshness signals and changes elsewhere on the web. Compare can support an experiment, but it cannot turn that experiment into causal proof.
7. The exports and documentation are refreshingly practical
Bing lets users export grounding queries, cited pages and time-series data to CSV or Excel, with active filters applied. That sounds basic. It is also what turns the report from something to admire in a browser into something an editorial or SEO team can actually investigate.
We also appreciate that Microsoft documents the data’s rough edges instead of pretending the dashboard is a perfect ledger. This is the part that belongs near the top of every serious review of the product.
Bing says the dashboard is a representative, aggregated view rather than a complete log of every citation. Low-frequency activity may not appear. Totals can differ between the page table, grounding-query table and timeline. The data is refreshed daily with a processing delay, and additional processing can refine the results.
There is an even stranger detail. When filtering a query and then viewing one of its pages, the citation count can differ from the count shown when filtering that page and viewing the same query. Microsoft says this can happen because the page and query views may be sampled over slightly different windows.
That’s it for now, let me know what you think about it and how have you used it so far. I’m expecting Bing team to gradually improve it, so we will make sure to update the report.
The post Bing Webmaster Tools AI Performance review: 7 things we love appeared first on Search Engine Watch.
