How to Track AI Citations with Bing Webmaster Tools

Here is the uncomfortable truth about most GEO tools, mine included: they ask an AI model what it thinks of your brand and hand you back a guess. Educated, useful, but a guess. Bing Webmaster Tools does the one thing none of them can - it reports what actually happened. Real citations, in real AI answers, to real users. First-party, logged, free.

And almost nobody opens it. The June 2026 update made that a bigger mistake than it already was.

Here is how I read the report now, what the new numbers actually mean, and the move I make off each one.

What changed in June 2026

The AI Performance report landed in public preview in February and is still preview - Microsoft has said it will sit there for months while it adds metrics, so expect more change. The June additions are the biggest yet: four new views, all in preview, that turn it from a vanity counter into something you can actually plan against:

Citation Share is the headline. If your site got 3 of the 10 citations Copilot handed out for a grounding query, your share is 30%. Bing now gives you a share-of-voice number natively, on its own surface, which it never did before. That one addition reframes the whole report.

You stop asking "am I cited" and start asking "what slice did I win, and who took the rest".

“

Microsoft previewed four new AI reporting features for Bing Webmaster Tools: citation share, grounding query-intent labels, grounding query topic labels, and Generative Engine Optimization (GEO)-focused recommendations.

First, the thing everyone reads wrong

The metrics that matter, and what each one is really telling you

1. Total citations. How many times your site was cited as a source across AI answers in the period. A count, not traffic - a citation is not a click, and Bing says so plainly. Treat it as reach, not revenue.
2. Average cited pages. The daily average of unique URLs of yours that got referenced. This is the spread number, and it quietly matters more than the headline. One page cited a hundred times is a liability; ten pages cited ten times each is a moat.
3. Grounding queries + intent. The engine's retrieval phrases, now tagged informational, commercial, or research. The intent tag is the cheat code - it separates citations that sit near a buying decision from the trivia you happen to rank for.
4. Citation Share + Compare. Your slice per query, and whether it moved versus the last 30 days. This is the first time Bing lets you watch yourself lose: your raw count can climb while your share falls because a competitor climbed faster. Direction over absolute, always.

What the report looks like in practice

The numbers below are illustrative - invented to show the shape, not pulled from any real account. But this is exactly the table you end up staring at:

Read it the way I do, and the story writes itself:

• You own the trivia. 64% share on "what is a gantt chart" feels great and pays nothing. Nobody buys software at the end of a definition.
• You own the head-to-head. 75% on "asana vs your-brand" is genuinely valuable - that is a buyer comparing, and the AI is handing them your side of the story.
• You are invisible where the money is. 5% and 6% on the two commercial and research rows. Forty buyers a period are asking the AI for the best tool for their use case, and you are in two of those answers.

That last line is the whole brief. The flattering numbers are a distraction; the two near-zero rows on high-intent queries are the work. A report that only showed total citations would have buried this - you would have seen a healthy count and gone back to sleep.

What to do when your Citation Share is low

A low share on a query that matters is not a mystery. It means the engine went looking for sources on that intent and yours did not make the cut - not clear enough, not fresh enough, not quotable enough. The fixes, in the order I run them:

1. Build the page around the grounding query's phrasing, not your keyword. The query is the engine's own words for the intent. If it grounds on "how to track team capacity" and your nearest page is titled "Resource Management Platform", you are forcing the model to bridge the gap. Answer the exact phrasing head-on, in an H2.
2. Lead with the liftable answer. Models cite the source that already wrote the sentence they want to say. Put a direct, quotable answer in the first 100 words, then support it - a summary line, an H2-per-question structure, a comparison table. Be the path of least resistance.
3. Refresh before it craters, not after. A share slide in Compare is a freshness warning you get two weeks early. Update the data, the date, the examples on the cited page the moment it slips, while you still hold the slot.
4. Make sure Copilot can even see you. Bingbot powers Copilot's index. I have watched sites indexed perfectly in Google Search Console and half-missing in Bing. Verify your domain in Bing Webmaster Tools and check coverage there, not just GSC. (More on the channel itself in why Copilot is the AI channel you are ignoring.)
5. Earn third-party mentions for the commercial queries. The "best X for Y" roundups, review sites, expert lists. On a head-to-head or "best tool" query, the model leans on what other people wrote about you, not what you wrote about yourself. One editor naming you beats a page of your own copy.

Your weekly fifteen minutes with the report

1. Sort grounding queries by intent, then by share. Commercial and research queries where your share is low are the highest-leverage gaps in the whole account - proven AI demand, and you are losing the slot.
2. Read each Topic cluster as a content brief. A cluster with citations but no dedicated page of yours is a page to write, with the demand already validated. You are not guessing at a keyword; the engine told you it needs a source here.
3. Run Compare every Monday. Overlay the last 30 days. A share drop on a query you used to own is the earliest warning you will get that someone out-published you - act before the count follows.
4. Treat foreign-language grounding queries as free market research. Queries in another language hitting your English page are a market signal. One native post, then measure - cheap probe, real answer.

The honest limits

Bing Webmaster only sees the Bing-grounded surface, including Copilot - not ChatGPT's full picture, not Perplexity, not Google AI Overviews. Its Citation Share is share of the Bing world, not the AI world. So it is half the dashboard, not the whole one.

The other half is cross-engine. GEOlikeaPro's SOV Dashboard measures your share against named competitors across ChatGPT, Perplexity, and Google AI, so the gaps Bing surfaces on one surface get pressure-tested across all of them. Use Bing for ground truth on Copilot; use cross-engine tracking for the rest.

Still - it is free, it is first-party, and it is the cleanest citation signal most teams have sitting completely unused. If you do one new thing in GEO this week, open it.

GEOlikeaPro turns citation tracking into a plan - cross-engine share of voice, the grounding-query gaps worth chasing, and the pages to write next. See how it works, free tier.