Back to Blog
Guides20 min read27 September 2026

How to Read Bing's AI Performance Report

A view-by-view guide to Bing's AI Performance report: what grounding queries, Citation Share and Intents mean, and what to change on your site.

By AI Schema Gen Team

Bing's AI Performance report is the only free report from a search engine that shows you which of your pages AI answers cite as a source, and the short search phrases the AI used to find them. Google's version tells you how often your site showed up inside its AI features. Bing's gets down to the page and the phrase.

That makes it useful, but also easy to misread. The numbers look like the rankings reports you already know, and they aren't. A citation count is not a ranking. A grounding phrase is not a keyword someone typed. A falling line on the chart does not come with a reason attached.

This guide walks through the report one view at a time and turns each one into a question you can answer and a decision you can make. If you're still deciding whether tracking AI visibility is worth your time at all, start with why track visibility, which covers when it pays off and when to skip it. This post assumes you've decided to look and want to know what to do with what you see.

Before You Start: Getting Into the Report

The report lives inside Bing Webmaster Tools, Microsoft's free dashboard for site owners. If you already use it, open your site and look for "AI Performance" in the left-hand menu. If you don't, setup takes a few minutes.

Add your site. The quickest route is to import it from Google Search Console. Microsoft's setup guide explains that sites imported this way "get automatically verified," so there is no separate proof-of-ownership step. You can also add a site by hand and prove you own it by adding a DNS record, uploading a small XML file, or adding a meta tag to your homepage.

Wait two days. The same guide says it "usually takes 48 hours" after a site is added before data appears. An empty report on day one is normal.

Know what it covers. Microsoft's help page for the report lists three places the citations come from: Microsoft Copilot, AI-generated summaries in Bing, and "select partner AI integrations." Microsoft doesn't name those partners on the page, so don't assume the report covers every AI assistant you've heard of. It covers Microsoft's own, plus some others.

Pick a date range. You can look at 7 days, 30 days, 3 months, or a custom range within the history Bing holds. For a monthly read, 30 days is the natural setting.

Know what's new and what's still a preview. The core report arrived in February 2026. Four newer views, called Intents, Topics, Citation Share and Compare, were added in June 2026 and are still described by Microsoft as a preview. Expect them to change, and expect some labels to be rough.

One Idea That Makes the Rest Make Sense

Before reading any number, it helps to understand what a "grounding query" is, because two of the six views are built on it.

When an AI assistant answers a question, it often searches the web first to find sources it can rely on. That step is called grounding. The short search phrases the assistant uses in that step are the grounding queries. Microsoft's February announcement describes them as "the key phrases the AI used when retrieving content that was referenced in AI-generated answers."

The important part is what they are not. The help page is direct about it: the phrases "are not full user questions or prompts." They are grouped, shortened versions of what the AI searched for. A person might ask Copilot a long, rambling question about whether they can claim their home office on their taxes as a freelancer. The grounding phrase that shows up in your report might be something much shorter, like "home office deduction self-employed."

This changes how you should use the list:

  • It tells you what the AI thinks the question is about. That's valuable. It's the AI's reading of the need, not the person's wording.
  • It is not a keyword list to paste into your pages. Some advice going around suggests taking each grounding phrase and working it into your copy. Be careful with that. The phrases are summaries the AI made after the fact, and the help page adds that they "may be determined differently across supported AI experiences and partners." Writing a page to match a summary is chasing a moving label. Writing a page that fully answers the underlying question is what got you cited in the first place.
  • It is a sample. Microsoft says the phrase data "represents a sample of overall citation activity." A phrase missing from your list doesn't prove the AI never used it.

Keep that in mind and the rest of the report becomes much easier to read.

A Monthly Read, View by View

The report has six main views. Rather than describe every button, here is the question each view answers and the decision it points to. Working through them in this order takes about twenty minutes once you're used to it.

1. Totals and the Timeline: Did Anything Break?

The top of the report shows total citations for the period, the average number of your pages cited per day, and a line chart of citations over time.

The question: is there a sudden drop or jump that lines up with something you did?

How to read it: look for sharp changes, not gentle drift. A steady line that wobbles week to week is normal. A cliff is worth investigating. Turn on Compare, which Microsoft describes as a way to "overlay a previous time period directly onto the current reporting view," and put this month next to last month. A drop that starts on the day you launched a redesign, changed your URLs or edited a key page is a lead.

The decision: if a drop lines up with a change you made, check whether that change broke something, such as a page that moved without a redirect, or a page accidentally set to be hidden from search engines. If nothing on your side changed, write the drop down and wait another month before acting.

The limit: Microsoft is clear that the chart can't tell you why something happened. The help page says trends "cannot be attributed to a specific cause or event" and lists causes as varied as changes in what people ask, updates to your content, and changes to the AI systems themselves. The timeline tells you when to look, never what to fix.

2. Pages: Which Pages Carry the Load?

The Pages view lists your URLs by how often each was cited during the period.

The question: are the pages being cited the ones you'd expect, and are important pages missing?

How to read it: sort by citations and look at the top ten. Two patterns matter:

  • A few pages doing all the work. It's common for a handful of pages to take most of the citations. That isn't bad, but it is fragile. If one of those pages changes or goes out of date, a large share of your citations goes with it. Treat your top cited pages like your top ranking pages: change them carefully and check afterwards.
  • An important page that isn't there at all. If your main service page, your pricing page or your best guide never appears, the first thing to check isn't the writing. It's whether AI systems can reach and read the page. A page blocked by robots.txt, set to noindex or built so its text only appears after scripts run can be invisible no matter how good it is. Our guide to whether AI crawlers reach your site walks through those checks.

The decision: protect the pages that carry the load. For missing pages, check access first, because no amount of rewriting helps a page the AI can't reach. If access is fine, the next suspect is clarity: a page whose opening never says plainly what it's for is easy to pass over.

The limit: the help page says the Pages view shows counts only and "does not indicate ranking, authority, importance, or a page's role within an individual AI answer." A page cited 200 times might be a minor footnote in each of those answers.

3. Phrase-to-Page Mapping: Is the Right Page Answering?

This is the most useful view in the report, and the one most guides skip. Microsoft added a link between the Grounding Queries view and the Pages view. In the words of Microsoft's advertising blog: "Select a grounding query to see its cited pages or select a page to see its associated grounding queries."

That makes it possible to check whether the AI is using the page you'd want it to use for a given question.

The question: for each phrase that matters to your business, which of your pages is the AI citing?

How to read it: pick five to ten phrases that describe what you actually sell or know about. Click each one and note the page it maps to. Then go the other way: click your most important pages and see which phrases lead to them. Three patterns come up:

  • The wrong page is being cited. A phrase about your service is mapped to an old blog post instead of your service page. The AI found the older post more useful for that question. Either the service page doesn't answer it clearly, or the blog post answers it better. Improve the service page, or link the post to it so the two work together.
  • Two pages share one phrase. Both pages get some citations for the same phrase. Sometimes that's fine. Often it means two pages are saying nearly the same thing and neither is clearly the best source. Consider merging them, or making each one clearly about a different part of the question.
  • A page is pulled into phrases outside its job. Your pricing page shows up for general "how does this work" phrases, for example. The AI is using it as a stand-in because no better page exists. Rather than editing the pricing page, give that question a page of its own.

The decision: each mismatch points at one specific page to edit, which is exactly the kind of result a report should give you.

4. Intents: Is the Page Shaped for the Question?

The Intents view sorts your grounding phrases into types of question. The help page lists fourteen labels: Informational, Media, Navigational, Commercial, Learn and Solve, Research, Live Event, Local, Comparison, Planning, Utility, Creation, Conversational and Other.

The question: what kinds of questions are your pages being used for, and does that match what you want?

How to read it: look at the split, then compare it to your business. A shop that mostly shows up for Informational questions but rarely for Commercial or Comparison ones is being used for background facts, not for buying decisions. A consultant who wants to be cited for Research questions but only appears for simple Informational ones may be answering the easy version of the question and leaving out the depth.

The decision: match the shape of the page to the type of question. Microsoft's help page gives examples of this itself: "comparison tables for Comparison, step-by-step guidance for Planning, or concise explanations for Informational queries." If you want more Comparison citations, a page that actually compares options side by side is more useful to the AI than a page that only describes your own product. If Informational questions are your target, short, direct answers near the top of the page help. Our guide to FAQs AI cites goes deeper on writing answers that stand on their own.

The limit: Microsoft notes that these labels come from its own automatic sorting and that "these insights can help inform content planning, but they do not guarantee citation outcomes." Treat an intent gap as a hypothesis to test, not a verdict.

5. Topics: Where Are You Thin?

The Topics view groups related phrases into broader subjects. Microsoft's June announcement gives the example of "solar panels," "solar energy efficiency" and "residential solar installation" all falling under one topic, Solar Energy.

The question: which subjects bring you citations, and which subjects you care about are missing or weak?

How to read it: list the topics in the report next to the topics your business actually covers. The interesting rows are the ones where you have plenty to say but few citations, and the ones where you're getting citations on a subject you've barely written about.

The decision: a topic with citations but thin coverage on your site is a good candidate for a new page or a fuller guide, since the AI is already looking to you for it. A core topic with no citations at all is a gap to fill.

The limit: Microsoft says plainly that during the preview, some labels "may still be broad," especially in specialised or niche fields. If your business is specialised, some topic groupings may be too vague to act on. Fall back on the phrase list when that happens.

6. Citation Share: How Big Is Your Slice?

Citation Share takes one grounding phrase and shows what percentage of all the citations for it went to your site. Microsoft's June announcement defines it as "the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query."

The question: for the phrases that matter most, is your share holding, growing or shrinking?

How to read it: the help page describes four patterns, and each leads somewhere different:

PatternWhat it suggestsWhat to do
High and stableYou're a steady source for that phraseKeep the page accurate and current. Don't rewrite what's working.
High but fallingOther sources are being cited moreCheck whether the page is out of date or less complete than it was.
Low on a phrase you care aboutYou're one of many sources, or a minor oneMake the page more complete, more current, or easier to follow.
GrowingYou're being cited more for that phraseNote what changed recently so you can repeat it.

The decision: focus on the "high but falling" rows first. They are pages that used to work and are slipping, which usually means something specific went stale.

The limit: Microsoft calls Citation Share "an observational metric," "not a ranking system or a competitive scoreboard," and adds that it "does not expose competitor domains, represent traffic share, or assign quality scores to content." You'll see how big your slice is, not whose slices make up the rest, and not whether anyone clicked.

A Worked Example

To see how the views fit together, picture a small accounting firm that offers tax preparation for freelancers and small businesses. This firm and its numbers are invented for illustration.

The timeline shows a steady month with no sharp drops. Nothing broke, so no emergency.

The Pages view shows the firm's blog post on home office deductions as the most cited page by a wide margin. The main "Tax Preparation for Freelancers" service page, the one that brings in clients, barely appears.

The phrase-to-page mapping explains why. Phrases like "freelancer tax filing help" and "self-employed tax preparer" map mostly to the blog post, not to the service page. Clicking the service page shows only a couple of phrases, both about the firm's name and address.

That's a clear diagnosis. The AI is looking for a page that explains how freelance tax preparation works and who can help with it, and the blog post happens to explain it better than the service page. The service page is mostly a list of services and a contact form.

The Intents view backs this up. Most phrases are Informational or Learn and Solve. Very few are Commercial.

The decisions:

  1. Rewrite the top of the service page so it plainly explains what freelance tax preparation involves, who it's for and what the process looks like, in a few direct sentences.
  2. Add a short question-and-answer section to the service page covering the questions the phrases point to.
  3. Link the blog post to the service page, so the two work as a pair.
  4. Leave the blog post alone. It's the page carrying the load.
  5. Note the date of the changes, and compare next month against this month.

Nothing in that list involves stuffing phrases into the page. Each change makes the service page a better answer to the question the AI was already trying to answer.

What the Report Can't Tell You

Every view above comes with a limit, and it helps to have them in one place before you draw conclusions.

  • You never see the actual answer. The help page says the data "does not show individual AI answers, exact prompts, or why a specific page was referenced." You know you were cited. You don't know what was said around the citation or whether it was accurate. Checking accuracy still means asking the assistants yourself.
  • Small numbers can disappear. "Very low or infrequent citation activity may not surface," according to the help page. A small site may see a sparse report even when it's being cited occasionally.
  • The views won't always add up. The help page warns that "totals may differ across views," for example between the Pages view, the phrase list and the timeline. That's expected, not a bug.
  • It's a sample. The phrase data in particular is described as a sample of overall activity.
  • It's one family of AI tools. It covers Microsoft's AI features plus unnamed partners. It is not a picture of every AI assistant.
  • It shows citations, not visits. Being cited and getting a click are different things. For visits that come from AI assistants, you need your analytics. Our guide to tracking AI traffic in GA4 shows how to separate them out.

None of this makes the report less worth reading. It means each number is a lead to check, not a finished answer.

How It Fits With Google's Report

Google launched its own generative AI reports in Search Console, and it's natural to want to line the two up. They measure different things, so they don't compare directly.

Google's report counts impressions: how often your pages appeared inside Google's AI features, such as AI Overviews and AI Mode. Bing's report counts citations: how often your pages were shown as a source in an answer, plus the phrases behind them. Google tells you that you were seen. Bing tells you that you were used, and hints at why.

Use them together rather than choosing one. Google's report is where most of your search audience is. Bing's report gives you the phrase-level detail Google's doesn't. Our breakdown of Google's generative AI report covers its counting rules and blind spots in detail.

Keep Your Own Record

The report keeps a limited history, with a three-month preset and custom ranges only "within available historical data." If you want to compare this autumn against next spring, you'll need your own copy.

The help page lists three downloads: the grounding phrases with their citation counts, the page-level citation data, and the timeline data for your chosen dates. Once a month, download all three into a folder named for the month. Alongside them, keep a short note of anything you changed on your site that month: pages rewritten, pages added, redirects, a redesign.

That note is what makes the Compare view useful months later. When a line moves, you'll be able to see what else happened at the same time instead of guessing.

Microsoft's own announcement also points to two related tools. If your content changes often, IndexNow lets you tell Bing right away when a page is added or updated, so the AI works from the current version. If you're a local business, keeping your listing on Bing Places up to date helps make sure your address, hours and contact details are correct when AI answers local questions.

Common Mistakes

Treating citations like rankings. A page with more citations isn't "ranking higher." Microsoft is explicit that the counts don't show importance or position within an answer.

Pasting grounding phrases into your pages. The phrases are the AI's short summaries, not your customers' words. Answer the underlying question well instead.

Reacting to one month. Citation patterns shift for reasons that have nothing to do with you. Look for changes that hold for two or three months, or that line up with something you did.

Rewriting your best-cited page. If a page carries a large share of your citations, change it carefully and check the effect. Improving a weak page is safer than reworking a strong one.

Fixing content when the problem is access. If an important page never appears at all, check that AI systems can reach it before you rewrite a word. Then check that its opening says plainly what it's for.

Reading Citation Share as market share. It covers one phrase at a time, doesn't name the other sources, and has nothing to do with traffic.

Forgetting to write things down. Without a note of what changed and when, the timeline and Compare views can only show you that something moved.

How to Check and Fix

The report is good at showing which pages AI is already using. It can't tell you why a page is being passed over. The usual reasons are the same ones that decide whether a page is cited anywhere: AI can't reach it, can't make sense of it, can't tell who's behind it, or can't find a clear answer in it.

A sensible order of work after your first read:

  1. Fix access for any important page missing from the Pages view.
  2. Fix mismatches found in the phrase-to-page mapping, one page at a time.
  3. Reshape pages where the intent split doesn't match what you want.
  4. Fill topic gaps with new pages where you have real expertise.
  5. Re-read next month and compare.

Steps one to three are checks you can run on your site today. Run a free readiness check to see whether AI systems can reach your pages, read them, tell who runs your site and answer questions from them, with a list of what to fix first.

FAQ

Is Bing's AI Performance report free?

Yes. It's part of Bing Webmaster Tools, which is free for any site owner who adds and verifies a site. The newer views added in June 2026 are described by Microsoft as a preview but are included in the same dashboard.

Why is my AI Performance report empty?

The most common reasons are timing and volume. Microsoft says data usually takes about 48 hours to appear after a site is added. Beyond that, the help page notes that very low or infrequent citation activity may not show up, so a small site with only occasional citations can see little or nothing.

What is a grounding query?

It's a short search phrase an AI assistant used to find sources before writing an answer, and one of your pages was cited as a result. It is not the question the person typed. Microsoft describes the phrases as grouped versions of what the AI searched for.

Does the report show citations from ChatGPT?

The report covers Microsoft Copilot, AI-generated summaries in Bing and what Microsoft calls "select partner AI integrations." Microsoft's help page doesn't list those partners, so don't read the report as covering any specific assistant beyond Microsoft's own.

Why do the numbers differ between the Pages and Grounding Queries views?

Microsoft's help page warns that totals may differ across views. One phrase can map to several pages and one page to several phrases, and the phrase data is a sample. Compare trends within a view, not totals across views.

Should I add my grounding queries to my pages as keywords?

Not as a rule. The phrases summarise what the AI searched for, and Microsoft says they can be worked out differently across its partners. Use them to understand the question, then make sure your page answers it clearly and completely.

Is your site ready for AI?

Get a free readiness score in under a minute. No signup, no card.

Run the free check