The questions where Bing’s AI cited someone else

Bing Webmaster Tools has had an AI Performance report since earlier this year: the pages of yours that Copilot and the AI answers in Bing cited, and the grounding queries behind them, meaning the searches the assistant ran before it wrote. On 13 October 2026 the report gained a tab called Topics Portfolio, in beta. The tab groups those grounding queries into topics and says three things about each: Topic Size, the approximate number of citations across all sites; Sites Cited, how many sites appear for the topic; and Queries, how many distinct questions it holds.

Open a topic and you get two tables. Your Grounding Queries are the questions on which your site was cited. Opportunity Queries are the questions in the same topic on which other sites were cited and yours was not. Under the topic table sits a cross-topic list of the top opportunity queries. You pick a window of one, two or three months.

Credit where it is due: nobody else publishes this. Search Console has no report of the questions behind AI answers, and the tools that sell AI visibility sample the engines from the outside. This is the engine’s own ledger, free, inside an account most sites already have.

Bing says the figures are directional signals rather than a precise ledger, and the tab’s design agrees. Topic Size and Sites Cited are approximate. The citation volume of an opportunity query is a range. The topic labels are written by a model and can be off. Topics with little activity are hidden, so a small site may see a short list or no tab at all: the tab appears only when there is enough data.

Three more things to know before you read a row as a task. First, language. On a Polish site the opportunity queries will include English, French or Dutch questions, because the assistant answers people everywhere and the topic is the same. Second, brands. A query naming a competitor’s product can sit in a topic as an opportunity, and it is not one. Third, absence. A query where you are not cited is not automatically a gap in your content. Often the page exists and the assistant could not or would not quote it: a snippet directive, a blocked crawler, a page that is empty without JavaScript, or simply a page that never states the answer in one sentence.

And one thing the tab is not: a measure of topical authority. It is one engine, one period, approximate, and it counts only the questions that engine happened to run. A good signal to plan with. A poor one to boast with.

An opportunity query is worth your time when it passes three checks, in this order.

• Language and business. Is it in the language your buyers use, and is it about something you sell or explain? If not, drop it and do not look back. The thumbs-down on the row tells Bing too.
• An existing page. Do you already have a page that answers this exact question? Look it up in your own crawl, not in your memory. In SilentStork the cluster map shows the pages a topic already has and whether they link to the hub; a site search of your own domain does the same, more slowly.
• A competitor’s brand. If the query names a product that is not yours, the assistant was right to cite them. Drop it.

What survives the three checks gets one of two verbs. A page exists: fix it. Usually the page talks around the question instead of answering it. Put the question in a heading, answer it in the first sentence under that heading, give the answer a date and a source, and check the three citability gates, because a page that cannot be quoted will not be, however good the answer.

No page: write it. Start from a brief that names the question, the heading that answers it and the two or three headings people also ask, and give the new page its links from the hub before it goes live. A page nobody links to is a page the assistant finds late, if at all.

Do this for ten queries, not a hundred. The list is three months of one engine’s behaviour, and the tab will be back next month with a new one.

The tables are small enough to read by hand. If you would rather have an assistant sort them, the prompt below does the three checks above and nothing more. It differs from the prompts going around in one respect: it asks for your list of pages first. Without it, a model will cheerfully decide whether you have a page on a subject, and the result reads like analysis while being a guess.

Paste your sitemap or the page list from your reading where the prompt asks for it. Keep the prompt in the language you want the answer in.

You are helping me read the Topics Portfolio (Beta) tab from Bing Webmaster Tools, AI Performance. The figures are approximate ranges from a beta, the topic labels are written by a model, and the queries can be in several languages.

My site: [address]
What the business does: [one sentence]
The language my buyers use: [language]
Pages I already have, with their titles: [paste the list from your sitemap or your reading export]

Below I paste the topics table (Topic, Topic Size, Sites Cited, Queries) and the opportunity queries, each with the topic it belongs to.

For every opportunity query, answer in a table:
1. Keep or drop. Drop it when the language is not my buyers’, when it names a competitor’s brand, or when it has nothing to do with the business. Say which reason.
2. For a kept query: name the page from my list that already answers it, or write “no page”. Never name a page that is not on my list.
3. If a page exists: what the page is missing to answer this exact question (the question in a heading, a direct answer in the first sentence, a date, a source).
4. If no page exists: a one-line brief with a working title, the question it answers and three headings.

Then list the five queries worth doing first, each with its reason in one sentence. Do not estimate difficulty or traffic: the data does not carry them. Where you need something I did not paste, say so instead of assuming.

[paste the data here]

We looked at building this into SilentStork and decided not to, for now, for three plain reasons. The tab has no export and no API, so the only way in would be pasting a file by hand. It appears only for sites with enough citations, and most of the sites we read are not there yet, so the screen would be empty for most of you. And its numbers are ranges with labels from a model, which is fine to read and poor to compute with: nothing built on them could meet the rule that every row in the product is a count of something that happened.

What the product does instead sits on either side of Bing’s tab. Citability tells you whether a page can be quoted at all, through three gates and a score you can take apart. The Citation Tracker asks an answer engine your own buyers’ questions once a week and counts, with no guessing, whether you were named and which page was cited, this week against last. The cluster map tells you which pages a topic already has. Bing’s tab fills the gap between them: the questions where somebody else was quoted.

If Bing adds an export or an API and the tab reaches the sites we read, there is one thing worth building: opportunity queries as a third source of pages to write, next to the searches you already show for and the ones your competitors win, with the same overlap check before any brief is written. Until then, the tab, the three checks and the prompt are the whole method.