Microsoft's AI Performance report in Bing Webmaster Tools expanded on June 16, 2026 with four preview features: Citation Share, Intents, Topics, and Compare. Citation Share is the one that changes how you measure AI search visibility. Instead of asking "did we get cited?", you can now see what percentage of all citations for a given query your site captured. This guide walks through the full workflow: reading the report correctly, segmenting it, separating your own work from platform noise, and turning the output into a repeatable weekly routine.
What you'll finish with
Who this is for: SEO and GEO practitioners, content leads, and site owners with a verified site in Bing Webmaster Tools who want a first-party measurement of AI citation visibility.
What you'll have when done: a correctly interpreted Citation Share baseline for your site, a segmented view of where you win and where you split citations, a 30-day Compare snapshot, and a prioritized list of content gaps to work next week.
Prerequisites:
- A site verified in Bing Webmaster Tools (free, Microsoft account, verification via XML file, meta tag, or DNS record)
- At least 30 days of accumulated AI Performance data; very new or very low-traffic sites may see sparse rows
- Access to your own analytics for the cross-check step
Time required: about 40 minutes for the first full pass, then roughly 20 minutes weekly.
Definition of done: you can answer three questions with numbers from the report — where am I cited, what share of each query's citation space do I hold, and which intent or topic gaps deserve content work next.
Before you start: access, account, and honest limits
Open bing.com/webmasters, sign in, and click AI Performance in the left navigation. The report covers citations of your pages inside AI-generated answers across Microsoft Copilot, Bing's AI experiences, and select partner AI surfaces. It defaults to the last 30 days, and you can export the data for spreadsheet work.
Know what this report is not before you build anything on top of it:
- There is no click data, so you cannot compute an AI click-through rate from it yet. Microsoft has talked about click reporting, but nothing had shipped as of mid-2026.
- Numbers are sampled rather than exhaustive, and the history window is short because the report only launched on February 10, 2026.
- There's no API either. Bing's team has said API access is planned for 2026, but it hadn't arrived by July. Manual export is the only path for now.
- Intents and Topics are AI-generated labels, and they're preview quality. Microsoft itself warns that some labels may be broad, especially for niche domains, and that potentially risky queries get filtered out of the classifications.
One more framing point. Microsoft describes Citation Share as "an observational metric, not a ranking system or a competitive scoreboard." Treat it as a trend line for your own site, not a leaderboard against named competitors.
Start with grounding queries, not the share number
The most common misread of this report starts at the query column. Those rows are grounding queries: the internal retrieval phrases Copilot's system generates when it searches Bing's index to assemble an answer. They are not what the user typed.
A single user question like "what's the difference between RPA and AI agents" can spawn three or four grounding queries, and only some of those retrievals turn into visible citations. Microsoft's own web search documentation confirms the engine generates a short search query informed by the user's prompt rather than passing the full prompt through.
So when you see a query string in the report, read it as "the phrase the retrieval system used to find citable sources," not "a keyword with search volume." A practical quality check: if a grounding query looks like odd fragments, that's normal. If an entire topic area you cover produces no rows at all, that's a signal your content may not be getting retrieved — check indexing and crawler access before touching the content itself.
Read Citation Share the way Microsoft intends it
Citation Share answers one question: of all citations shown across all sites for a specific grounding query, what percentage went to your site? If an answer pulls from ten sources and two are yours, your share for that query is 20 percent. The math is simply your citations divided by total citations for that query.
Here's a real view of the report, with the Citation Share column highlighted:
The interesting lessons live in the ratio between citation count and share. Look at two rows from this table:
Pattern | Citations | Citation Share | What it means |
|---|---|---|---|
High citations, tiny share | 5.4K | 0.26% | You're in the mix but crowded out. Many sources split this query's citation space. Volume without ownership. |
Moderate citations, high share | 979 | 22.16% | Fewer total appearances, but you dominate this query's citation space. This is what "owning" a query looks like. |
Low citations, high share | 590 | 8.87% | A niche query where you're a primary source. Small but defensible territory. |
A page with 50 citations at 10 percent share is losing to eight other sites. A page with 20 citations at 60 percent share essentially owns that query in Copilot answers. Same citation count, completely different strategic picture — which is exactly why raw citation totals were never enough.
A quick quality check: sort the report by Citation Share descending, then separately by Citations descending. If the same pages top both lists, you have genuine strongholds. If the two lists barely overlap, you're spread thin, and consolidation should outrank expansion this month.
Cut the data by Intents and Topics
Raw query rows tell you where you appear. The two classification layers tell you why.
Intents label each grounding query by what the underlying AI interaction was for: Informational, Commercial, Navigational, Learn and Solve, Research, Creation, Local, and others. This matters because citation behavior differs sharply by intent. E-commerce publishers often show up strongly in comparison and shopping interactions; educational publishers show up in research and learning ones. If your citations cluster in Informational while your revenue depends on Commercial queries, you're visible where it's cheap and invisible where it pays.
Topics group related grounding queries into thematic clusters. "Solar panels," "solar energy efficiency," and "residential solar installation" fold into a Solar Energy topic. Reading by topic instead of by query reveals two things query rows hide: emerging areas where the AI systems treat you as an authority, and coverage gaps where a cluster cites you thin or not at all.
In the screenshot above, notice how one site's citations split across Translation Technology & Tools (11.91 percent share), Technology (up to 34.70 percent), Travel (15.78 percent), and AI Chatbots & Assistants (5.64 percent). That spread is the point: Topics let you see your authority map the way the AI systems see it, concept by concept rather than keyword by keyword.
Recovery path if the labels look wrong for your domain (in preview, sometimes they are): don't build strategy on the mislabeled rows. Microsoft states these classifications are maturing and will improve as data accumulates. Re-check after a few weeks, and in the meantime trust your own grouping of the exported query list.
Use Compare to tell your work apart from platform noise
Compare overlays a previous period onto your current view (for example, the last 30 days against the prior 30 days) so you can see how citation activity moved after a content update, a seasonal shift, or a change in the AI ecosystem itself. Write-ups from teams testing the feature describe per-row comparison columns (Last, Prev, and Diff) alongside the overlay view.
Before you celebrate any spike, do this check: match the jump's date against Bing's release timeline. The report launched February 10, 2026, and the four new features rolled out June 16, 2026. Independent analysis by PinMeTo found that some early-June jumps traced to Microsoft backfilling historical data, which a Microsoft PM described as "regular data backfilling with no anomalies." A step change right at those dates deserves skepticism before you write a case study about your own brilliance.
The clean version of a Compare cycle:
- Pick a window where you shipped something specific (a content refresh, new section, IndexNow pinging habit).
- Set Compare to that window versus the equivalent prior window.
- Look at Diff on the specific queries your change targeted, not the site-wide total.
- Attribute only what you can explain. A number you cannot explain is not a number you can act on.
Turn the report into a priority queue
The report becomes useful when it produces a worklist. Score each topic cluster on two axes: current Citation Share and business value. Then work the matrix:
High business value | Low business value | |
|---|---|---|
High share | Defend. Keep fresh, keep indexed, watch Compare monthly. | Leave alone. |
Low share | Attack first. These are citations you're splitting with others where winning more share pays. | Optional. Only worth content work if the cluster is growing. |
Concretely, low-share, high-value clusters usually respond to the same moves that earning citations has always responded to: put a direct 40–80 word answer at the top of the section, phrase headings as the questions people actually ask, add verifiable statistics with attribution, mark the page up with schema, and ping IndexNow on every update so Bing's index has the current version. If the index lacks your page, Copilot cannot cite it regardless of quality.
Cross-check one thing outside Bing's dashboard before committing budget: your own analytics referrer data. AI assistant referrals (from ChatGPT, Perplexity, Gemini, Claude, and Copilot) are small but real, and comparing their trajectory against your citation counts keeps the two honest. If you want a second read on visibility across AI platforms beyond Bing's ecosystem, Auspia's AI Search Visibility Checker covers the same question from the outside in.
A 20-minute weekly loop that keeps the numbers honest
Once the first pass is done, the maintenance routine is short:
- Open AI Performance and set the date range to the trailing 30 days. (2 min)
- Scan Citation Share on your top 10 queries by citations. Note any share that moved more than a few points. (5 min)
- Check Compare against the prior period. If anything moved a lot and you can't explain it, check Bing's release notes before celebrating. (3 min)
- Skim Intents and Topics for new clusters appearing or old ones vanishing. New clusters are early authority signals; vanishing ones are freshness problems. (5 min)
- Update the priority queue with one cluster to attack and one to defend this week. (5 min)
Expect GEO effects to show up in this report within two to four weeks of a change, consistent with what Princeton's GEO researchers observed. Slower than rankings, faster than backlinks.
FAQ
Is Citation Share the same thing as keyword share of voice? It's the AI-era analogue, not an equivalent. Share of voice measured your slice of impressions or clicks on a SERP. Citation Share measures your slice of the citations inside an AI answer, where there may be no click at all. The strategic question it answers is similar: not "am I present?" but "how much of the available visibility did I capture?"
Can I see which competitors hold the rest of the citation space? No. Microsoft explicitly says the metric does not expose competitor domains. You see your percentage and the total citation count for a query; who took the remaining share stays invisible.
Why did my citations jump suddenly in June 2026? Two things happened at once: Microsoft rolled out the four new features on June 16 and, per PinMeTo's analysis, backfilled historical data. If your jump lands near that date or the February 10 launch, attribute cautiously.
Does a high Citation Share guarantee traffic? No. The report has no click data, and AI answers can satisfy users without a visit. Treat citation share as a visibility and influence metric, and validate against referral traffic in your own analytics.
How is this different from Google's equivalent? As of mid-2026, Google folds AI Overview data into its broader Search Console reporting with no per-URL citation counts, and its dedicated AI report has been thinner and regionally limited. Bing's AI Performance report remains the most concrete first-party AI citation data any major search engine publishes.
Author: Jasper Quinn, AI Search Product Researcher at Auspia. Jasper tracks feature changes across search and AI platforms and writes practical walkthroughs for measuring AI search visibility.
