There is a question everyone in marketing has been asking for two years and nobody could answer cleanly: how often does an AI draw on my content as a source — and how often on my competitors'? Visibility in generative answers was estimated, inferred, extrapolated. Agencies ran prompt panels, scraped answers and sold the result as measurement.
Microsoft closed that gap in 2026 in two steps. The AI performance report has been in Bing Webmaster Tools since February. On 16 June 2026 four further functions arrived — among them Citation Share, the first metric from a major platform that expresses your visibility not as an absolute number but as a share. All free, and according to Microsoft globally in preview.
1. Why AI visibility used to be guesswork
Classic SEO has a dependable measurement chain: ranking, impression, click, conversion. Generative answers had none of it. When Copilot or an AI overview answers a question and names three sources, the site owner saw a referral in their analytics tool at best — and usually not even that, because many answers work without any click at all.
The consequence: the whole field of Generative Engine Optimization (GEO) worked with proxy data. You tested prompts, logged which brands were named, and extrapolated. That is better than nothing, but it is not measurement — it is a sample with an unknown population. Deriving budget decisions from it meant standing on thin ice.
2. What Microsoft shipped
The AI performance report launched in February 2026 in public preview with four base metrics: Total Citations (how often your website was cited overall), Average Cited Pages (how many of your pages on average), Grounding Queries (the AI's internal retrieval phrases) and Page-Level Citation Activity (which individual page was drawn on how often).
The June update is additive: it lays four analytical lenses over that base instead of replacing it.
3. Citation Share: your slice of the citation pie
Citation Share answers a question that absolute counts cannot: of all the sources an AI answer drew on for this query — how many were mine? The calculation Microsoft gives is as simple as it sounds: your own citations for a grounding query, divided by the total number of citations shown for that query, times 100.
An example: an AI answer draws on ten sources, three of them from your website. Your Citation Share for that query is 30 per cent.
Microsoft is unusually clear in framing it: Citation Share is an observational metric — not a ranking system and not a competitive scoreboard. It names no competitor domains, does not represent a share of traffic and does not judge content quality. That clarification is not a footnote but load-bearing: a high Citation Share means visibility, not automatically revenue. It belongs next to your own analytics data, not in its place.
4. Grounding queries are not user queries
The most important conceptual point in the whole report is also the one most often misreported. The Grounding Queries metric shows the search phrases the AI system formulated itself in order to retrieve content for its answer. It does not show what a human typed.
When somebody asks Copilot a long, colloquial question, the model breaks it into one or more more precise retrieval phrases. Those internal phrases are what land in the report.
Microsoft's own recommendations from February point in the same direction: strengthen depth and expertise, improve structure with headings, tables and FAQs, back statements with evidence, keep content current and reduce ambiguity. At heart that is good editorial work — only this time with a metric you can measure it against. What that looks like day to day is described in our article on the AI-supported editorial workflow.
5. Intents and topics: numbers become a priority list
Citation Share gets the headlines, but only the three other functions make it actionable. Intents sorts grounding queries into more than seven intent categories — Microsoft names informational, commercial, navigational, learn and solve, research, creation and local among them. The use is diagnostic: for the first time you see which intents win you citations and which ones you simply do not appear for.
That is the moment a number becomes a task. An online shop with strong informational citations and near-zero visibility in the commercial bracket has a concrete, closable gap — and it only becomes visible through this breakdown.
Topics groups related queries automatically into thematic clusters. Microsoft's own example bundles "solar panels", "solar energy efficiency" and "residential solar system" into a "solar energy" cluster. Microsoft openly admits the labels can still be coarse in preview — particularly in specialised niches.
Compare overlays an earlier period on the current view, for example the last 30 days against the 30 days before. This function has a dependency that governs the whole timetable: it is only useful once history exists. And history cannot be created after the fact.
6. The honest limits of the report
Honest measurement means naming the limits too. The Bing report is currently the most transparent native AI visibility analysis from a major platform — and it leaves real gaps open.
On top of that comes the market question any honest article has to ask: Bing holds a single-digit share of the global search market, and estimates vary widely. So the report is not a representative picture of all AI answers. Its value lies elsewhere — it is currently the only free, platform-native metric of its kind, and the mechanics of generative answers are similar enough across systems to derive working hypotheses. As a complement, the Google Search Console generative AI report remains the second pillar.
7. A baseline in 30 minutes
The practical part is pleasingly short. Anyone who has never opened Bing Webmaster Tools needs less time to get started than for a reporting meeting: verify the property, open the AI performance report, note the current Citation Share and export the grounding queries. That is the baseline. Everything after it is a comparison.
8. What this means for smaller companies
The temptation is to dismiss the topic in one sentence: "Bing is small, so it does not matter." That falls short, for three reasons.
First, the demand side is already there. A Pew Research Center survey from February 2026 among 5,119 US adults found that around 60 per cent of respondents read the AI summaries at the top of search results and about half use AI chatbots at all — considerably more than in 2024. The figures are from the USA and do not transfer one to one, but the direction is unambiguous.
Second, Copilot is embedded where B2B decision-makers already work: in Windows, in Edge and in Microsoft 365. For companies with services that need explaining, that is not a sideshow but a channel their audience is already using.
Third — and this is the real argument — getting started costs nothing but half an hour, while waiting gets expensive. Compare only works with history, and history cannot be created retroactively. Set a baseline today and you have a trend line in November. Start in November and you have an empty report in November.
Also worth reading: Generative Engine Optimization: becoming visible where AI answers and E-E-A-T and AI authority: why ChatGPT cites you of all people.
9. FAQ: the key questions about Citation Share
What is Citation Share in Bing Webmaster Tools?
Citation Share is a metric Microsoft introduced on 16 June 2026 in the AI performance report of Bing Webmaster Tools. It shows your percentage share of all citations an AI answer drew on for a particular grounding query. The calculation: your own citations divided by the total number of citations from all pages for that query, times 100. If your website supplies 3 of 10 citations, your Citation Share is 30 per cent. Microsoft describes the metric explicitly as observational and not as a ranking system.
What is the difference between a grounding query and a user query?
A grounding query is the search phrase the AI system formulates internally to retrieve suitable content for its answer. It is not what a human typed. When somebody asks Copilot a long, colloquial question, the model breaks it into one or more cleaner retrieval phrases. Those internal phrases are what the report shows. Treating the two as the same means reading the data wrongly: it is not a keyword list, but a view of how the model classifies your content.
Does the AI performance report cover ChatGPT and Google AI Overviews?
No. The report covers citations in Microsoft Copilot, in Bing's AI summaries and in selected partner AI experiences. ChatGPT, Perplexity and Google AI Overviews are not included. That makes the report a precise window into the Microsoft ecosystem, but not a universal dashboard for AI visibility. For the other channels you still need Google Search Console, your own analytics data and third-party tools.
Does the AI performance report cost anything, and is it worth it for small companies?
The report is free and, according to Microsoft, available globally in preview. The only requirement is a verified property in Bing Webmaster Tools. For smaller companies it is interesting above all because the comparison view needs history: data cannot be produced retroactively. Set a baseline today and in three months you have a dependable trend line. The effort is about half an hour.
