Imagine this situation: your website performs well in the traditional Google results list, yet something still feels missing from the picture. When you ask Microsoft Copilot about your field of expertise, not a single sentence in the answer cites you. Then there is the opposite case as well: one professional subpage regularly appears as a source in AI-generated answers, while the number of clicks barely moves. Confusing, isn’t it?

These two phenomena show exactly why it is no longer enough to think only in rankings. Generative search engines — whether Bing, Copilot or their partner surfaces — do not simply return a list of results. They merge the content of several sources into one direct answer and attach citations to it. Organic traffic simply cannot show this type of visibility.

This is where Bing AI Performance enters the picture: first-party data showing when, in what context and how frequently an AI system cited your content. Microsoft launched the Bing Webmaster Tools AI Performance report in public preview in February 2026, then expanded it on June 16, 2026 with new features: intents, topics, Citation Share and period comparison. The report shows aggregated citation data from Bing, Microsoft Copilot and certain partner AI surfaces in one place.

To interpret this dataset correctly, however, you first need to separate four concepts: number of citations, the set of cited pages, grounding queries and Citation Share. In this article, we do exactly that — step by step, with practical examples and a small interactive calculator at the end that lets you immediately test your own data.

For a deeper look at the differences between classic SEO, AEO and GEO, we have also created a separate guide: A practical framework for AI SEO, GEO and AEO.

1. What is the Bing AI Performance report?

AI Performance is a separate report inside Bing Webmaster Tools. It does not examine classic search rankings; instead, it shows how and to what extent a website becomes a source in AI-generated answers. In other words, it does not measure where you rank in the list, but whether you enter the small set of sources the AI actually uses.

It is therefore worth moving away from the reflex that “the higher the number, the better.” The real value of the report lies in the patterns it reveals: which pieces of content, in which topics, and how often become reference points. For a broader understanding of this AI visibility concept, we recommend this overview: What does AI visibility really mean?

2. Cited pages: which content does AI consider a usable source?

The truly important question is not how many pages are indexed on the domain. What matters is how many of those actually enter the source set of AI answers. A 500-page online store may have only four or five pages that are genuinely “citable” — and that is exactly the number worth watching.

Examining the list of cited pages answers several questions:

It is important to distinguish between two similar-sounding metrics. “Average cited pages” is a site-level average: how many unique URLs appear as sources per day in aggregate. “Page-level citation activity”, on the other hand, shows citation counts assigned to specific URLs. Microsoft also emphasizes that these metrics are not authority scores or rankings — they are meant for observing citation patterns, not for grading pages.

For more on technical accessibility — in other words, why an AI bot may not “see” your content at all — read this article: Why is content marketing invisible to AI bots?

3. What are grounding queries?

The Hungarian phrase for “grounding query” can be described as a source-grounding retrieval query, but since the English term is increasingly used in professional language, we keep both concepts in this article.

Put simply: a grounding query is the keyword phrase or search formulation the AI system used when it retrieved the web information needed to support its answer. According to Microsoft’s definition, these are the important phrases with which the AI found the content that was later cited — and the displayed data represents a representative sample of total citation activity.

It is not the same as the user’s question

This is the most important and most commonly misunderstood point. A grounding query does not necessarily match what the user actually typed into Copilot or Bing.

This decomposition shows exactly how AI breaks one complex user question into smaller, independently answerable information tasks. Once you understand this, you can adapt your own content planning to the same logic — instead of building around a single keyword, you should cover an entire question space.

Analyzing grounding queries can therefore reveal:

• which question areas the AI considers the website relevant for;
• which terminology it associates with the content;
• whether it connects the content to informational, commercial, local or research intent;
• which related topics are worth creating new content for;
• where there is a gap between the intended positioning and the topics the AI actually perceives.

The latest version of the Bing report also groups grounding queries into broader intent categories — such as informational, commercial, navigational, research, local, creative or problem-solving categories — which gives an even more detailed picture of how content fits into the user journey.

You can read more about how source selection, authority and content structure work together here: How do source selection, authority and content structure work together?

4. What does Citation Share mean?

Citation Share shows what percentage of all displayed citations for a given grounding query came from your website.

Let’s look at a concrete case: in AI answers connected to a specific grounding query, 100 source citations appeared in total. Of these, 18 citations pointed to your domain, while 82 belonged to other sources. In this case, your Citation Share is 18 percent.

Why is this more valuable than citation count alone? Because it adds context. Twenty citations can be an outstanding result in a narrow professional subtopic, but insignificant in a topic area where thousands of citations are generated daily. Citation Share shows exactly this relative weight.

At the same time, Microsoft clearly indicates that this is an observational metric — nothing more.

Citation Share also moves continuously: user interest, model updates, content freshness, partner data sources and changes in the entire web source pool all affect it. This is not a system error — it is the nature of generative search.

You can read more about Citation Share and the logic of measuring AI citations on our homepage: AI SEO and AI citation measurement.

5. The three metrics must be interpreted together

The metrics introduced so far tell only half-truths when viewed separately. Only when they are placed next to one another does the real picture of a domain’s AI visibility emerge.

A simple rule of thumb is useful:

• citation count shows activity;
• number of cited pages shows content breadth;
• grounding queries show semantic context;
• Citation Share shows relative presence;
• period comparison shows direction.

With Bing Compare, for example, you can compare the current 30 days with the previous 30 days, or compare two custom periods. However, caution is important: a shift is not necessarily the result of a single optimization action — model updates, competitor movement and demand seasonality can all shape the numbers.

6. Practical Bing AI Performance analysis workflow

You can read more about the relationship between classic search engine optimization and search marketing here: Search marketing and search engine optimization.

7. How can the number of cited pages and Citation Share be increased?

You can find a full presentation of the optimization process covering technical SEO, GEO and AEO here: AI visibility guide: technical SEO, GEO and AEO.

8. The most common interpretation mistakes

Summary: AI visibility is becoming a measurable business process

The most important strength of Bing AI Performance is not that it provides one final AI visibility score. Quite the opposite: through several separate signals, it shows where, in which topics and at what relative share your content is being used to support AI answers.

Cited pages show which pieces of content actually become sources. Grounding queries reveal how AI interprets the website’s topics. Citation Share puts appearance counts into context — showing what share your own domain holds in the competition for a given topic.

The future SEO question is therefore no longer only “where does the page rank in Google?” — but just as much:

When AI systems answer which questions, do they rely on our website?

Frequently Asked Questions