Which of your pages Microsoft Copilot already cites, and why its own free report is the first place to look
Bing Webmaster Tools has an AI Performance report in public preview that names the pages of yours used as sources in Copilot answers and the grounding queries that pulled them in. Microsoft publishes exactly what it does not tell you as well, and that boundary is what makes it usable. No other operator in this category publishes anything equivalent.
Published by AI Knows Us · Updated 20 October 2026
Microsoft already tells you which of your pages Copilot uses as sources, and most people working on this have not opened the report. Bing Webmaster Tools carries an AI Performance report, in public preview, covering "Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations". It gives page-level citation counts, visibility trends over time, and the grounding queries, which Microsoft describes as the "key phrases the AI used when retrieving content that was referenced in AI-generated answers". That makes Copilot the one surface in this category where the work starts with your own first-party data rather than with a guess.
What the report gives you, in Microsoft's own words
- Total Citations, how often your content appears as a source in AI answers, which Microsoft says does not indicate placement.
- Average Cited Pages, defined as "the average number of unique pages from your site that are displayed as sources in AI-generated answers per day".
- Grounding queries, the phrases the system used when it retrieved your content.
- Page-level citation activity, counts for individual URLs.
- Visibility trends over time, and since October 2026 a Topics Portfolio in beta inside the same report.
And the limits, which Microsoft states itself rather than leaving you to discover: Average Cited Pages "does not indicate ranking, authority, or the role of any page within an individual answer", and "The data shown represents a sample of overall citation activity", with a note that the grounding query metric will be refined as more data is processed. Read the trend rather than any single day, and treat the query list as a strong sample rather than a complete log.
The distinction that decides how you use it
The report tells you your content was used as a source. It does not tell you whether Copilot named your business in the answer, or named somebody else. Those are two different findings with two different counts, and they fail for different reasons: a page can be cited while the recommendation goes to a competitor, and a business can be named from a directory listing while none of its own pages are cited at all.
So pair the report with your own reading. Open Copilot, ask your five real buying questions with your brand name absent, and write down who it names. Then open the report and see whose pages did the work. The two lists together are the brief; either one alone is half of it.
What to do with grounding queries, which is the part worth the hour
Grounding queries are the closest thing in this trade to being told what the machine wanted. Three passes over that list, in order:
- Find the queries where a weak page of yours is doing the work. A thin page pulled in for a commercially important phrase is the best possible signal: the surface already wants an answer there and you are supplying a poor one. Improve that page before writing any new ones.
- Find the queries you did not expect. The phrases will not match your keyword list, because they are retrieval phrases rather than things people type. Phrases that recur are subjects to cover properly, and they often read like clauses rather than searches.
- Find your most-cited page and ask what shape it is. Then check whether your other important subjects are published in that shape. On this surface the cited pages are usually the specific ones: a definition, a comparison, a number with a date on it, a limit stated plainly.
Write down the date you pulled the list and the citation count beside it. Everything you claim later about improvement rests on having a dated first reading, and the report gives you one for nothing.
The technical half, and the one control to check first
Microsoft reports Copilot citations through Bing Webmaster Tools alongside AI-generated summaries in Bing, which tells you where the plumbing runs: the work that gets you indexed and crawlable for Bing is the work that puts you in range of this surface. Three things are worth half an hour each.
- Verify the property in Bing Webmaster Tools, because the report exists only for verified sites. Plenty of teams verified Search Console years ago and never did this one, and the citation data has been accumulating without anybody reading it.
- Check your snippet controls. Microsoft has added support for the
data-nosnippetattribute, so a passage wrapped in it is a passage you have asked to have left out. Find out whether anybody on your team has used it on a page you want quoted. - Submit changes rather than waiting. IndexNow exists precisely so a changed page is picked up quickly, which matters here because the thing you are testing is whether a revision changes an answer, and a slow pickup confuses the result with the delay.
What the report cannot do, and what to buy if you need it
Three gaps, and they are the three things a paid tool sells on this surface. The report covers your own property, so there is no competitor column. It reports the queries that reached you, so it cannot show the buying questions where you are absent, and absence is what changes a publishing plan. And it counts citations rather than storing the answer text, so you cannot read what Copilot actually said.
If you need those, the vendors that read Copilot on a plan with a published price are the place to look, and we are not among them: our own five surfaces are ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, and Copilot is not one of them. Rankscale publishes the deepest Copilot reporting we found, including its local and map pack results, sponsored ad share of voice and shopping cards. Otterly carries "MS Copilot" in the base four of every published tier. Scrunch carries it in the four on its entry tier. The pair of vendors the assistants name together most often differ on exactly two engines, one of them this one, on a page that carries no durable inbound link from anywhere on this site: 147 of our live pages in this category have none, counted across 2,054 stored bodies.
Questions about getting cited and named by Microsoft Copilot
Do I need a tool to see my Copilot citations?
No, and that is unusual in this category. Verify your site in Bing Webmaster Tools and the AI Performance report gives you page-level citation counts and grounding queries at no cost. Buy a tool when you need the competitor column, a prompt set you chose, or the text of the answer.
Is the data complete?
Microsoft says the grounding query data "represents a sample of overall citation activity" and the report is in public preview. Use it for direction and for comparing one month against another, and keep your own dated readings alongside it for anything you intend to quote to a client or a board.
Does a citation mean Copilot recommended me?
It means a page of yours was used as a source. Microsoft is explicit that the metrics do not indicate placement or "the role of any page within an individual answer". Whether you were recommended is a separate question answered by reading the answer, and the two counts should be reported separately and never averaged.
What kind of page gets cited here?
On this surface the pattern that shows up is specificity: a page that answers one narrow question with a concrete, checkable statement. The grounding queries tell you which narrow questions, for your site in particular, which is better than any general guidance including this paragraph. Pull your own list and follow it.
Why does my Copilot citation count move so much week to week?
Because citation depends on what people asked that week as much as on what you published. Read the trend across a quarter and compare months rather than days, and note alongside it when you shipped changes, so a rise has a candidate cause rather than a shrug.
Does any other operator publish something like this?
Not for the surfaces in this category. Google documents that sites appearing in its AI features are "included in the overall search traffic in Search Console" and "reported on in the Performance report, within the Web search type", with no separate line for AI Overviews or AI Mode. So on Google's surfaces the measurement has to come from reading the answers, while on Microsoft's one half of it arrives from the operator.
Start with the report you already own
Of the eight surfaces people ask about in this trade, exactly one hands a site owner its own citation data, per page, with the phrases that pulled it in, for free. Log in, pull the grounding queries, find the three where a thin page of yours is already carrying commercially valuable retrieval, and rewrite those three properly. Then ask Copilot your own buying question and see whether the answer now names you rather than merely using you. That sequence costs an afternoon, needs no vendor, and gives you a dated starting number that every later claim can be measured against.