Bing's AI Performance Report: See the Exact Queries Copilot Uses to Find You

Bing Webmaster Tools now shows the exact search queries Microsoft Copilot used to ground its answers in your content. That report is called AI Performance, and it is the first time a major search engine has handed site owners a direct list of the prompts behind an AI citation instead of a traditional click. If you have ever wondered what someone actually typed before Copilot named your brand, this report is where you find out.

Every AEO practitioner runs into the same wall. You can see traffic. You can see rankings. You cannot see what happens between a person’s prompt and the answer an AI model writes back. The AI Performance report opens a window into that gap, at least for Bing and Copilot. It will not tell you everything about your AI visibility, but it tells you something no other major webmaster tool currently offers: the literal query behind the citation.

This matters most if your buyers already treat Copilot as a shortcut past traditional search. Enterprise buyers researching software, homeowners comparing local contractors, and B2B teams vetting vendors increasingly ask Copilot the question directly instead of scanning ten blue links. Every one of those conversations runs through a grounding step, and until this report existed, you had no way to see your side of it.

What the AI Performance report actually shows

The AI Performance report lives inside Bing Webmaster Tools, next to the standard Search Performance report you likely already check for organic queries. Instead of showing clicks and impressions from people typing into Bing’s search box, it isolates the queries Copilot ran against the web while building an answer, then flags the ones where your site’s content ended up inside that answer.

Each row represents one grounding query. You see the query text itself, how many times Copilot grounded on your site for that query during the reporting window, and which specific page or pages got cited. Some rows will look familiar, close variations of keywords you already target in Bing and Google. Others will look nothing like your keyword research, because they are phrased the way a person talks to an AI model, not the way a person types into a search box.

The report also breaks activity down by date, so you can see whether grounding on a given page is a one time spike tied to a single trending topic or a steady pattern building over weeks. That distinction matters more than the raw count. A page that grounds ten times in one day because of a single trending topic is a different signal than a page that grounds ten times spread evenly across a month.

What Copilot grounding queries actually are

Grounding is the step where an AI model reaches outside its training data to pull in current information before it answers. Copilot does not answer every question purely from memory. For anything that depends on current facts, specific businesses, or recent content, it runs a retrieval step first, searches the web, and feeds relevant pages into the answer it is about to write. The query it runs during that retrieval step is the grounding query.

A grounding query is not always identical to what the user typed. Suppose someone asks Copilot, “who does AEO work for HVAC companies in Phoenix.” Copilot might break that into several underlying searches: one for AEO agencies generally, one for HVAC marketing specifically, one for Phoenix based services. Any one of those underlying searches could surface your site, and the AI Performance report shows you that underlying search, not the original conversational question a person actually asked.

This distinction changes what you optimize for. You are not only trying to match how people phrase questions to Copilot. You are trying to match the searches Copilot itself generates when it decomposes those questions, which often land closer to traditional keyword phrases than the original prompt does. It is one more reason AEO and SEO overlap more than most people expect, even as the two disciplines diverge in other places.

The technical prerequisites Copilot needs before it can ground on your site

None of this data shows up if Copilot cannot reach your site in the first place. Grounding depends on the same technical foundation that determines whether any AI crawler can access, parse, and trust your pages.

Start with crawler access. Copilot’s retrieval step relies on the Bing index, which Bingbot builds by crawling your site the same way it always has. If your robots.txt file blocks Bingbot, or blocks it from the specific pages you want cited, those pages cannot ground on anything, regardless of how well written they are. This is worth checking directly rather than assuming, since plenty of sites inherit a restrictive robots.txt template without anyone noticing.

Schema markup matters here too, for the same reason it matters across every AI answer engine. Clean Organization, Article, and FAQPage schema helps Copilot understand what a page is about and who published it, which supports both the initial grounding decision and the citation that follows. A page with no schema and a page with full schema coverage can cover the identical topic and still get treated differently by a retrieval system trying to decide which source to trust.

Submission speed plays a role too. Bing’s IndexNow protocol lets you push new and updated URLs directly to Bing the moment they publish, instead of waiting for Bingbot to discover them on its own schedule. A freshly published page that answers a grounding query cluster will not show up in the AI Performance report until Bing has actually crawled it, and IndexNow is the fastest way to close that gap.

How to access the AI Performance report in Bing Webmaster Tools

Getting to the report takes a few minutes once your site is verified. Here is the full path from a fresh login.

1. Verify your site. Add and verify your domain in Bing Webmaster Tools if it is not connected already. No report, including AI Performance, will populate without verification.
2. Open the Reports section. From the left navigation inside Bing Webmaster Tools, open Reports.
3. Select AI Performance. Choose the AI Performance report from the list. This is the report that separates Copilot grounding queries from standard web search queries.
4. Set your date range. Cover at least 30 days so you are reading a pattern instead of a single noisy day, and apply any available filters for query type or page.
5. Review the query list. Read the grounding queries sorted by volume, and note which pages got cited for each one.
6. Export and log the data. Pull the report into a spreadsheet or your existing citation tracking, so you can compare grounding query volume month over month instead of treating each visit to the report as a one time snapshot.

If your site is new to Bing Webmaster Tools, expect a lag before this report shows anything. Bing needs to crawl your site, index it, and have Copilot actually cite it during a reporting window before a single row appears. That lag is diagnostic on its own. A verified site with strong Bing organic rankings that still shows zero grounding activity after a month or two is telling you something real about how Copilot is treating your content. The report’s refresh speed is rarely the actual explanation.

Reading the report without drowning in query noise

A raw list of grounding queries is not an insight by itself. Once real data starts flowing, three habits turn the report into something you can act on.

Sort by grounding volume, not clicks

Clicks and grounding measure different things. A query can ground on your site dozens of times without producing a single click, because Copilot answered the question directly and the person never needed to visit your page. Sorting by grounding volume first shows where your content is doing the most work inside AI answers, even when that work never shows up anywhere else in your analytics.

This is also why grounding volume alone should not replace your other metrics. A page with high grounding volume and no resulting traffic might be doing exactly what you want, informing an AI answer without needing a click, or it might mean Copilot is citing you while sending all the resulting interest to a competitor named alongside you. The report tells you the grounding happened. It does not tell you what the reader did next.

Split branded queries from unbranded queries

Grounding queries that already contain your brand name confirm something you likely knew already, that Copilot can find you when someone asks about you directly. The unbranded queries deserve closer study, because those are the moments Copilot chose your content over a competitor’s without being told to look for you by name. Pull those into a separate list and treat that list as your real AEO scoreboard.

Track which pages get named

Match every grounding query back to the specific page Copilot cited. A pattern usually appears fast. A handful of pages account for most of the grounding activity, and most of the site contributes nothing at all. That pattern shows where your citation-worthy content actually lives, and it is rarely the pages a team assumes are the strongest.

Watch for seasonal and trending spikes

Grounding volume is not flat throughout the year. A page about open enrollment periods, tax deadlines, or seasonal maintenance will show a spike in grounding activity around the relevant dates and near silence the rest of the year. Do not read a spike as a permanent gain, and do not read the quiet months as a loss. Compare the same period year over year once you have enough history, the same way you would with any other seasonal traffic pattern.

Why Bing shipped this before Google or OpenAI

Google Search Console still reports on classic organic search: impressions, clicks, and average position by query. It does not currently break out which queries triggered an AI Overview citation versus a standard blue link. ChatGPT and Perplexity offer no webmaster facing report at all. Bing is the outlier here, and the reason has more to do with its existing infrastructure than any particular willingness to share data with site owners.

Copilot runs on the same Bing index and the same Bing Webmaster Tools platform that has existed for over a decade. Microsoft did not need to build a new measurement system from nothing. It needed to add a new report on top of a pipeline that already tracked queries, indexing, and site verification. That existing plumbing is the real explanation for why Bing could ship AI specific reporting well ahead of Google, which runs Search and AI Overviews across a more fragmented set of internal systems, or OpenAI, which has no organic search index or webmaster relationship to build on in the first place.

Common patterns in early AI Performance data

After walking multiple sites through their first look at this report, a few patterns show up often enough to expect them going in.

Most sites see nothing at first. A newly verified site, or a site with thin Bing organic visibility, frequently shows zero grounding queries for the first several weeks. That is not a broken report. It reflects a site Copilot has not yet found reason to cite.

Grounding concentrates on a small number of pages. Once data starts appearing, it rarely spreads evenly. A handful of pages, often not the homepage, tend to account for most of the activity. These are usually the pages that answer one specific question clearly, rather than pages trying to cover an entire topic broadly.

Branded queries show up before unbranded ones. Early grounding activity tends to skew toward queries that already include the brand name. Unbranded grounding, where Copilot cites a site without being prompted with its name, tends to build later, as entity signals and content depth catch up.

Grounding and clicks move independently. A page can gain grounding volume for months without any visible change in Bing organic traffic, because being cited inside an AI answer and earning a click are two separate events. Do not expect this report to track your existing traffic dashboards.

Grounding queries reveal which competitor content Copilot trusts instead. When a grounding query fails to cite you at all, the same query often surfaces a named competitor across a handful of manual test prompts. Logging those near misses next to your own grounding data turns the report into a competitive gap list, not only a self assessment of your own pages.

Bing’s AI Performance report versus every other way to measure AI citations

The AI Performance report is not the only way to find out whether AI models cite your brand, and it should not be the only one you rely on. It covers exactly one platform. Here is how it compares to the other methods teams use to track AI visibility.

The report’s biggest advantage is that Microsoft hands you the data directly, with no querying, no logging, and no guesswork required. Its biggest limitation is that it only covers Bing and Copilot. If your buyers spend more time in ChatGPT or Perplexity, this report tells you nothing about those platforms. For a full picture across every engine your buyers actually use, pair it with a dedicated methodology like the one we cover in our guide to AI citation tracking tools, which walks through manual and automated options for ChatGPT, Perplexity, and Google AI Overviews.

Turning grounding queries into a content plan

Once you have a few months of grounding queries logged, the report becomes a content brief you did not have to write yourself. Copilot is telling you, in its own words, what it searched for before it decided your site deserved a citation.

Suppose your report shows three grounding queries clustered around “how to switch marketing agencies without losing momentum,” each citing the same page on your site, plus a fourth query, “marketing agency contract cancellation terms,” that grounds on a page mentioning cancellation in a single sentence. That fourth query is the clearest signal in the batch. Copilot wants an answer on cancellation terms badly enough to cite a page that barely covers it, which means a dedicated section, or a new page entirely, is likely to catch more of that grounding activity going forward.

1. Group grounding queries by topic. Cluster the unbranded queries into themes. A page that grounds on five related queries is a stronger candidate for expansion than one that grounds on a single isolated query.
2. Compare grounding queries to your existing content. For each cluster, check whether the cited page actually answers the grounding query directly, or whether it got cited despite covering the topic only in passing. The second case is your highest priority rewrite.
3. Build new pages around clusters with no page behind them. If a cluster of grounding queries has no obvious page addressing it, that is a content gap Copilot has already told you is worth filling.
4. Write FAQ sections that mirror the grounding query phrasing. Grounding queries sit close to the language AI models actually search for. Working that phrasing into FAQ questions and headings makes future grounding more likely, not less.
5. Recheck the report the following month. Confirm whether the pages you updated picked up new grounding activity. This is the most reliable feedback loop available for whether a specific content change increased Copilot citations.

Where the AI Performance report runs out of road

Treat this report as one input, not a full measurement system. It has real limits, and knowing them keeps you from over trusting a single data source.

• It only covers Bing and Copilot. ChatGPT alone handles a large share of AI search volume in most categories, and this report has nothing to say about it. Perplexity, Gemini, and Google AI Overviews sit outside its scope entirely.
• It shows grounding activity, not sentiment. A grounding query tells you Copilot pulled your page into its retrieval step. It does not tell you whether the resulting answer described your brand favorably, mentioned a competitor in the same breath, or buried your citation below other sources.
• It leans on your existing Bing visibility. Copilot relies heavily on the Bing index for retrieval, so a site with weak Bing organic performance will typically show weak AI Performance data too. The report reflects your Bing foundation as much as it reflects any AI specific work.
• It reports with a lag. Grounding activity does not appear in real time, which makes the report better suited to trend analysis over weeks than to monitoring a single citation event as it happens.
• It will not tell you why a query stopped grounding. If a grounding query disappears from one month to the next, the report has no explanation field attached. You are left to compare your own content, technical, and competitor changes against the drop yourself.

Because of these limits, the report belongs inside a broader tool stack, not on its own. Our breakdown of the best AEO tools in 2026 covers the tracking, auditing, and citation monitoring tools worth combining with what Bing gives you for free.

Fitting grounding query data into a full AEO audit

Grounding query data answers a narrow question well: what did Copilot search for before it cited me. A full AEO audit answers a wider question: why does or does not AI cite me at all, across every engine that matters to my buyers.

Use the AI Performance report as one data source feeding that larger audit, not as a replacement for it. If your grounding queries reveal five pages doing most of the work, a proper audit checks whether those same five pages carry the schema markup, crawler access, and entity signals that would help them earn similar citations on ChatGPT and Perplexity too. If your grounding query list is thin or empty, an audit checks whether the problem is content, technical access, or entity authority, before a content sprint gets built on a guess.

Our 12 point AI visibility audit checklist walks through exactly that structured process, and it is built to run alongside whatever platform specific data you already have, including the AI Performance report.

A 30 day plan for your first month of data

If you are starting from zero, here is a realistic first month.

1. Week 1. Verify your site in Bing Webmaster Tools if it is not already, and confirm the AI Performance report loads. Do not expect meaningful data yet if your site is new to the platform.
2. Week 2. Pull whatever grounding queries have accumulated so far, even a short list. Separate branded from unbranded queries and match each one to a cited page.
3. Week 3. Pick your two or three strongest grounding query clusters and rewrite or expand the pages behind them, working the grounding phrasing directly into headings and FAQ content.
4. Week 4. Set a recurring reminder to pull this report monthly, and log the numbers somewhere you will actually revisit. A report nobody looks at again is worse than no report at all.

Four weeks will not produce a dramatic shift in grounding volume. What it will produce is a baseline, a short list of pages worth watching, and a habit of checking data that most competitors in your category are not checking at all.

Getting a second opinion on your grounding data

Reading a spreadsheet of grounding queries is the easy part. Knowing which clusters deserve a content sprint, which pages are quietly underperforming despite decent grounding volume, and how Bing’s numbers compare to citation rates on ChatGPT and Perplexity takes a wider view than any single report can give you.

AEO Hunt folds Bing’s AI Performance data into the same AI visibility and AEO work we run across every platform that matters to your buyers, so grounding queries stop being an isolated data point and start feeding an actual content roadmap.