Instrument Review
What does the Bing Webmaster Tools AI Performance report actually tell you?
Microsoft is the first platform to hand site owners first-party AI citation data. It's genuinely useful - and it answers a different question than the one most brands are asking.
28 Labs · August 2026
The AI Performance report, added to Bing Webmaster Tools on February 11, 2026, shows how often your pages were cited in AI answers across Microsoft Copilot, Bing's AI summaries, and select partner surfaces. It counts citations - it doesn't show the answers, doesn't cover ChatGPT, and can't tell you whether the AI recommended you or the competitor it quoted you next to.
What is the AI Performance report?
It's a free dashboard inside Bing Webmaster Tools for any verified site. Microsoft shipped it on February 11, 2026, then expanded it on June 16, 2026 with four preview features: Intents, Topics, Citation Share, and Compare. It's the first time a major AI answer provider has published its own citation data to site owners, which makes it both a useful instrument and an official confirmation that AI answer visibility is now a reporting category of its own.
Total citations
How many times your site appeared as a source in AI-generated answers over the selected period.
Average cited pages
The daily average of unique URLs from your site referenced across Microsoft's AI experiences.
Grounding queries
Sample search phrases the AI generated internally to retrieve your content (definition below).
Page-level citations
Citation counts per URL - which of your pages the AI actually reads.
Citation Share (June 2026)
Your site's percentage of all citations for a given grounding query - the closest thing to share-of-voice in the report.
Intents, Topics, Compare (June 2026)
Intent buckets (informational, commercial, research), thematic clusters of the queries you're cited for, and period-over-period comparison.
What are grounding queries?
Grounding queries are the search phrases an AI assistant writes for itself when it needs live web content to answer a user's question. They're not the user's prompts. A user asks one question; the engine silently fires several reformulated searches, reads the winning pages, and builds the answer from them. Your page gets cited when it wins one of those internal searches.
This is the most valuable data in the report, because it's Microsoft showing you the machine's side of the conversation - the queries you actually have to win, in the engine's own words. It's also the report's built-in blind spot: you see what the engine searched for, never what the human asked.
Which AI surfaces does it cover?
Microsoft surfaces only: Copilot, Bing's AI answer summaries, and select partner integrations. That's a meaningful slice of AI answers, but note what's absent:
- ChatGPT isn't covered. It launched search on Bing's infrastructure, but OpenAI now blends its own crawler (OAI-SearchBot) and index with third-party providers. Bing indexation still helps you in ChatGPT - it's one of the doors - but ChatGPT citations don't appear in this report.
- Gemini, Claude, and Perplexity aren't covered, and each retrieves differently.
What it can and can't tell you
The report shows
- How often Microsoft's AI surfaces cite your pages
- Which pages get read, and for which grounding queries
- Your share of citations per query (Citation Share)
- Intent and topic patterns, and movement over time
The report can't show
- The answer text - what the AI actually said about you
- Whether you were named or recommended, or a competitor was
- What real users asked (prompts stay hidden)
- Anything from ChatGPT, Gemini, Claude, or Perplexity
- Clicks, traffic, or revenue from being cited
Microsoft says it plainly: the metrics reflect citation frequency, not ranking, prominence, or how a page contributed to an answer.
Being quoted isn't being picked
Here's the distinction that decides whether this data helps you commercially. A citation means the engine read your page while building an answer. It doesn't mean the answer recommended you. An AI answer can cite your comparison page and still send the buyer to someone else - your content did the work, the competitor got the customer.
So citation counts measure your performance as a source. They say nothing about your performance as an answer - whether you're the brand named when a buyer asks "who should I use for X?" That second question is the one that moves revenue, and no citation dashboard answers it. You get it by asking engines real buyer questions, on a schedule, and reading what they say - the approach we cover in what AI visibility tools can and can't measure.
How to use the report well
1
Verify your site in Bing Webmaster Tools
2
Read the grounding queries you're cited for
3
Check Citation Share on queries that matter
4
Pair it with answer-level measurement
Free, first-party, ten minutes to set up. If your business isn't verified there yet, that's this week's job.
- Treat grounding queries as intelligence, not vanity. They tell you which questions the engine already brings your pages into - and by omission, which commercial questions you're absent from.
- Watch Citation Share on your money queries. Being one of eight sources is different from being the source.
- Don't optimize for citations alone. Rising citation counts with no presence in the actual answers means you're feeding the machine, not winning it.
- Remember the coverage boundary. This is your Microsoft mirror. Your ChatGPT, Gemini, and Perplexity reality can look completely different.
The honest readUse the free instrument - genuinely. Then be clear about where its vision ends: one vendor's surfaces, citations without answers, machine queries without human questions. If your AI visibility strategy starts and ends with this dashboard, you're managing the part of the problem Microsoft chose to show you.
What we do about it
We measure the answer layer: a frozen set of real buyer questions, asked across ChatGPT, Claude, Gemini, and Perplexity every wave, with every count re-derivable to specific questions and answers. First-party citation data like Bing's slots in as corroborating evidence - it's the engine's own receipt that it reads what we publish. The two layers together tell you what a citation count alone never will: whether AI actually picks you.