GEO | 2026-10-25 | 8 min read
Bing AI Performance: Track Citations and Find Gaps
Understand Bing AI Performance, Topics Portfolio, and citation share. Use a worked example and worksheet to choose which content to improve.
Direct answer: Bing AI Performance reports citations to your website across supported Microsoft AI experiences. Use its cited pages and topic opportunities to investigate missing answers, then decide whether to improve an existing page, create a distinct guide, or skip the question. Citation share measures relative citation activity within its stated scope; it does not measure every AI platform or your sales.
Written by: Esmail Hanif, AI Visibility Strategist & Founder, Martecks
What Bing AI Performance tells you
A customer may read an AI answer that cites your website without visiting it. A normal traffic report will miss that exposure. Bing AI Performance provides a view of source usage: citations, cited pages, and grounding queries across Copilot, Bing AI summaries, and selected partner experiences.
Microsoft introduced the report in February 2026. It measures a defined set of experiences, so use it alongside other evidence in your AI search strategy. Treating one provider's report as a census of AI search would give you a misleading baseline.
Open the report and choose a period
In Bing Webmaster Tools, select your verified site and open AI Performance. Record the property and reporting period before comparing numbers. Keep a copy of the observations you use for a decision, whether that is an available export or a dated note.
Start with a page you care about commercially. For a local service business, that might be the page explaining an important repair rather than the blog post with the most broad-interest traffic. Ask what the cited page helps someone decide.
Sources: Open Bing Webmaster Tools
What Topics Portfolio adds
Announced on October 13, 2026, Topics Portfolio groups your grounding queries into themes. Topic Size estimates citations across sites in a theme; Sites Cited and Queries estimate its breadth. Opportunity Queries highlight gaps in your recorded presence. These are directional citation signals, not keyword search volumes.
Microsoft calls this a beta. Classification can be imperfect, branded queries may appear as opportunities, and smaller sites may lack enough data to see the view. Begin with a relevant topic rather than automatically chasing the largest one.
An example: turn a gap into a content decision
Imagine a drain-repair company reviewing three relevant questions. The following scenario is hypothetical. It is a decision exercise, not a screenshot or measurement from a customer account.
The company already has an emergency repair page, but the page never explains its real attendance hours or call-out process. That is an omission to fix. Publishing a second page with almost the same purpose would leave customers choosing between two incomplete answers.
Repair versus replacement is different. A useful answer may require inspection findings, warning signs, cost factors, and cases where repair is unsuitable. That can justify a separate guide if the company has the expertise to explain it.
| Question under review | Existing evidence | Decision and reason |
|---|---|---|
| Emergency drain repair availability | Relevant service page with missing practical details | Update: fill the omission on the page customers need |
| Drain repair or replacement? | No explanation of how the decision is made | Create: a distinct decision guide with an expert-reviewed example |
| Discount code for another named provider | No relevant offer or relationship | Skip: a reported opportunity need not fit the business |
Inspect the evidence before choosing the fix
A gap is a reason to investigate. It does not establish why your page was absent. Check whether a suitable page exists, whether it is accessible, and whether it answers the actual decision. Then check what evidence the answer requires.
If the question is about reliability, another self-description may add little. Relevant reviews, verified credentials, and documented work could be more useful. An incorrect profile needs a correction. A real service weakness needs operational attention. Citation engineering explains how to work with those outside sources.
A manual AI check can help you inspect an answer and its sources, but it may not reproduce the reported event. Record it as a separate observation rather than pretending it reconstructs the original user's question.
Use the decision worksheet
Download the citation opportunity worksheet and keep one row per finding. It separates the reported observation from the action you choose. The example row is labelled hypothetical; replace it with your own evidence.
Record the question or topic, report period, existing URL, missing evidence, decision, owner, and follow-up date. Leave an outcome blank until you have observed it. A completed article is a shipped action, not proof that citations improved.
Use the query fan-out tool when you need ideas for questions around the same decision. Its generated suggestions are planning hypotheses, not measured Bing queries or search volume.
Check whether the action helped
Save the page change date and the original observation. At the next review, compare periods with similar duration and reporting coverage. Examine both citations and business outcomes; one may move without the other.
Suppose a fixed manual sample contains 40 valid answers and your brand appears in 12. Its observed mention rate is 30%. That invented example describes the sample only. It cannot be called 30% of all AI answers, nor can it be compared directly with Bing Citation Share.
For brand share of voice, define the denominator separately. One option is your brand appearances divided by all appearances of a fixed set of brands in the collected answers, counting each brand once per answer. Publish that definition with the result. Changing the competitor set changes the metric.
Use an AI visibility scorecard to keep the measures together. If a report improves after an edit, record the sequence; model changes and demand can also affect the result. A before-and-after chart alone does not isolate the cause.
Why is the Topics Portfolio tab missing?
Microsoft says the view depends on sufficient site presence and ecosystem activity. A missing tab therefore does not establish that no AI system has ever cited you. Avoid making technical changes solely to force a report to appear.
While evidence is limited, keep a small manual tracking sample and check the technical access of your priority pages. Our guide to tracking without expensive software explains the collection process.
Are grounding queries the exact questions people asked?
Microsoft describes grounding queries as retrieval phrases associated with cited content. Treat the displayed phrase as a clue to the information need. It is not a complete conversation transcript or a guaranteed copy of the customer's wording.
For example, a phrase about repair costs does not tell you whether the person was comparing providers, checking a quote, or researching a future job. Read your enquiry history before deciding which of those needs deserves a page.
Does this replace a paid AI visibility tracker?
It may cover an important reporting need without another subscription. It does not establish coverage of every ChatGPT, Gemini, Google, or other AI answer. Check the stated coverage of each report and product you use.
Paid tools can add scheduled questions, geography, competitor monitoring, exports, and team workflows, depending on the plan. Compare AI visibility tools only after identifying what your existing reporting cannot answer. Buying more collection does not automatically create better evidence or better content.
Choose one useful correction
Open a topic that matters to a buyer and inspect one question. Find the page that should help, identify the missing evidence, and assign one change. If the question does not fit your business, leave it alone.
A clear decision with a recorded reason is more useful than a long list of topics nobody will act on.
Sources: Microsoft: AI Performance launch, Microsoft: Citation Share and Compare, Microsoft: Topics Portfolio beta