Bing Webmaster Tools AI Performance is a free report showing which of your pages AI answers cite, and which searches triggered those citations. Microsoft shipped it into public preview in February 2026 and expanded it in June with Citation Share, Topics, Intents, and a Compare view. It's the closest thing available to first-party data on whether AI systems use your content, it costs nothing, and most site owners have never opened it.
That last part is the opportunity. While the industry sells AI visibility dashboards built on simulated prompts, Microsoft quietly shipped observed citation data for free.
What does the AI Performance report show?
Six things, and it helps to know which are load-bearing.
Citations count how often your content appeared as a cited source in an AI answer. Pages cited tells you which URLs earned them, which is often the surprise: the page you'd expect to be your AI front door frequently isn't.
Grounding queries are the short phrases the AI used when retrieving and citing your content. These are not full user prompts, and the distinction matters. A person asks a messy paragraph-long question; the system decomposes it into terse retrieval phrases. Those phrases are what you're seeing, and they're the most valuable column in the report. We wrote up the mechanism separately in what grounding queries are.
Topics cluster those queries into themes, so "solar panel cost" and "residential solar pricing" roll up together. Intents classify them (Informational, Commercial, Local, Research, and others), which is how you separate the citations that sit near revenue from the ones that don't. Compare overlays the current period against the prior one.
Then there's Citation Share: your slice of all citations shown for one grounding query. Three of ten citations reads as 30 percent.
What the report will not tell you
Four limits, and you should hold all of them before you read a single number.
It doesn't measure clicks. The report tells you that you were cited, never whether anyone came. That's a real gap, not a nitpick, because citation without traffic is the defining pattern of AI search right now. Pair it with the GA4 and Search Console work in measuring AI search traffic or you'll be optimizing a number with no revenue attached.
It doesn't show position or prominence. Being the first source in an answer and being the fourth both count as one citation.
The data is sampled, and Microsoft has said it may be refined retroactively as more gets processed. Small numbers will move on you. Don't build a narrative on a query cited four times.
It covers Microsoft's surfaces: Copilot, Bing's AI answers, and select partner integrations. Not Google's AI Overviews, not Perplexity, not ChatGPT directly. ChatGPT has long drawn on Bing's index alongside OpenAI's own crawlers, which makes this a defensible proxy for retrievability, and a proxy is not the thing itself. Anyone selling you Bing's report as "your AI visibility" is overselling it.
How do you turn it on?
Fifteen minutes, no cost, no sales call.
1. Add the site (5 min)
bing.com/webmasters, sign in, add your domain.
Fastest path: "Import from Google Search Console."
Your verified GSC properties import already verified,
no DNS edit, no file upload. Bing re-checks ownership
against GSC periodically, so if you ever revoke that
access, add a meta tag or DNS record as backup.
2. Submit the sitemap (2 min)
Sitemaps > add your sitemap.xml URL. Same file you
gave Google.
3. Turn on IndexNow (5 min)
Most CMS platforms have a plugin or setting. It pings
participating engines the moment you publish or update.
Google does not participate, so this is a Bing-side win,
which is exactly the side that feeds Copilot.
4. Open AI Performance (3 min)
It needs a few days of data before it says anything
useful. Set a calendar reminder for a week out rather
than refreshing it tonight.
Setup verified August 2026. Microsoft ships changes to this tool often, so trust the interface over any guide, including this one.
How do you read Citation Share without fooling yourself?
Sort every grounding query into four buckets. This is the read we run for clients, and it's yours to copy.
OWNED ~40%+ share. You're the default source.
Job: protect it, and check the page converts.
CONTESTED ~10-40% share, real citation volume.
Job: lift it. Biggest wins live here, because
you're already retrieved. Not a standing start.
UNDER-REPRESENTED Under ~10% on a high-volume query.
AI knows the topic and barely uses you.
Job: work out what the cited sources answer
that you don't.
ABSENT Topic or Intent with activity, you're nowhere.
Job: coverage gap. Publish or expand.
Rank by opportunity, not by share. Opportunity is citation volume multiplied by intent value, which means a contested commercial query beats an owned informational one nearly every time. Commercial, Local, and Learn-and-Solve intents sit closest to money.
Then read Compare for direction. Share falling on a query you used to own is more urgent than any new query you might chase, and it's the signal most people miss because a falling number still looks like a number.
The highest-value move the report enables is unglamorous: mismatch detection. When grounding queries phrase things differently than your pages do, they say "cost" and you say "pricing," they say "for small teams" and you say "for SMBs", add their language to the page that's already being cited. A wording fix on a retrieved page beats a new page, and it compounds immediately. The same logic drives the tactics in how to get cited by ChatGPT.
The monthly loop
Thirty minutes, once a month. Quarterly is too slow for a surface this young; weekly is noise given the sampling.
1. Export grounding queries with citation counts and share.
2. Bucket them: owned, contested, under-represented, absent.
3. Sort by volume x intent. Take the top three.
4. Pick one move each:
LIFT add the sub-answer the query implies
ENTER publish or expand for an absent Topic
DEFEND refresh a page whose share is falling
HARVEST make an owned commercial page convert
5. Log the date, the query, the move, the share.
Next month, compare. That log is your evidence.
Step five is the one everyone skips and the only one that turns a dashboard into a practice.
Do you need a paid AI visibility tracker?
Probably not yet, and it costs us to say so.
Most paid AI visibility tools work by running prompts on a schedule and recording which brands appear. That's simulation: useful, genuinely, for surfaces with no first-party reporting, which is still most of them. But it's inference about what an AI might say, while Bing's report is a record of what got cited. Observed beats simulated when you can get it, and here you can get it for nothing.
So the order is: open the free report, run the loop for a quarter, learn which of your pages AI systems reach for and which language they use. If you then need coverage across Google's AI surfaces and Perplexity, buy the tracker with a real question in hand instead of a subscription in search of one.
We run this loop as part of the Assay phase of the Reforge Method, alongside Search Console and GA4, because AI citation data belongs next to the search data rather than in a separate dashboard nobody opens. The KPI stack it feeds is laid out in how to measure AI search visibility.
The blunt version: a free report from Microsoft will teach you more about your AI visibility this quarter than most paid tools will, and the only thing standing between you and it is fifteen minutes of setup.