Bing AI Performance helps you understand how your website is cited in supported AI experiences. Use it to identify the pages and subjects receiving attention, then decide whether those pages are accurate, useful and relevant to your business. A citation is evidence of visible sourcing, not a click, a ranking position or a sale.
The current report also goes beyond its original launch metrics. Microsoft introduced Intents, Topics, Citation Share and Compare in a June 2026 preview update. This guide reflects the official material available at the time of writing, including those additions, and explains how to use the evidence without turning a sampled reporting view into a claim about the whole AI market.
Start with the content question you need to answer
Choose a practical question before reviewing totals. Which service guides are being used as sources? Are citations concentrated around subjects the business still wants to support? Is an old page receiving attention despite containing information that needs review? Those questions can lead to useful work even when the numbers are modest.
A citation can matter for several reasons. It may introduce a reader to your explanation, support a comparison or help an answer describe a process. The report does not automatically tell you which of those outcomes happened for a particular customer. You need to connect the page’s purpose with the available reporting context.
For a hypothetical supplier of commercial kitchen equipment, a cited maintenance guide may be useful to owners researching service requirements. It does not necessarily indicate demand for a new appliance purchase. A report that treats all citations as sales opportunities would miss that distinction and could push the content team towards the wrong objective.
Keep the business question visible in the report. “Check whether the maintenance guidance remains current” is an actionable response. “Increase the AI score” is much less useful unless the score’s meaning and its relationship to a decision are clearly explained.
Know which experiences and units are included
Microsoft’s original AI Performance announcement describes coverage across Microsoft Copilot, Bing AI summaries and selected partner integrations. It introduces total citations, average cited pages, grounding queries and page level citation information. That coverage should not be described as every assistant, every brand mention or every AI answer on the web.
Keep the unit beside the chart. Total citations and average cited pages answer different questions: one concerns citation activity and the other the breadth of pages appearing over time. A website can have concentrated activity around a small number of useful pages without that concentration being a problem.
The current AI Performance help documentation describes sampled, aggregated data with a processing delay. Sparse activity may be absent, and totals can differ between views. Use the report for directional analysis within its scope rather than exact accounting of every answer.
Record the property, dates and filters with any export. If somebody later asks why the page table differs from a chart, you need the original view to investigate. A spreadsheet detached from its filtering context can create apparent inconsistencies that are really differences in the question being asked.
Read grounding queries as summaries of context
Grounding means using source material to support an answer. In this report, grounding queries are grouped descriptions of that context rather than a transcript of individual customer prompts. A short phrase can therefore represent a broader set of questions. Do not treat it as an exact sentence that someone typed.
The practical value is thematic. A query group can suggest that a page is being used in discussions about cost, process, comparison or another relevant subject. Inspect the associated content and ask whether it answers that kind of question accurately. Avoid assuming that every citation came from a buyer at the same stage of a decision.
For the hypothetical equipment supplier, a grouped phrase about cleaning schedules could lead the team to inspect its maintenance guide. The review might find that the guide clearly distinguishes manufacturer instructions from general planning advice. That is a useful quality check even if the exact wording of the original user questions remains unavailable.
Do not create a new article for every grouped phrase. Several phrases may relate to the same useful explanation, and a single existing page may serve them well. A content decision should be supported by customer needs and a genuine information gap, not merely by the existence of another row in the report.
Use Intents and Topics without treating labels as verdicts
The June 2026 preview update adds Intents and Topics. Intents classify query context, while Topics group related grounding queries. Microsoft notes that the classifications can be broad, particularly in specialised subjects. Keep the preview qualification and inspect the underlying content before accepting a label as a precise description of customer intent.
These views can help an owner move beyond a long list of short phrases. A topic grouping may reveal that several useful pages relate to one business capability. An intent label may suggest that your content appears in research or comparison contexts. Both are starting points for interpretation rather than instructions to rewrite the site.
Consider a hypothetical industrial service page labelled under a broad informational category. The business should not conclude that the page has no commercial role. Buyers often need technical information before requesting a quote, and the report’s label does not reveal every person’s motivation.
Use your operational knowledge to assess the fit. Ask the people who handle enquiries whether the grouped subjects reflect real customer questions. If the classification seems too broad, retain the official label in the evidence record and add your explanation separately. Do not alter the label in a way that makes your interpretation look like a platform finding.
Understand Citation Share before comparing percentages
Citation Share describes your site’s proportion of citations for a particular grounding query. Microsoft says it is observational, does not identify competitor domains and is not a traffic share or content quality score. It should not be presented as your business’s universal share of AI recommendations.
A hypothetical calculation shows why counts and shares can move differently. Suppose a site receives twelve of sixty citations in a defined illustrative pool, giving it a 20 per cent share. In another period it receives fifteen of one hundred, giving it a 15 per cent share. Its citation count rose while its share fell. These invented numbers explain the arithmetic; they are not a real report or a claim about expected performance.
The business interpretation requires context. The subject might have attracted more activity overall, or the distribution of sources might have changed. A falling share alone does not prove the page became worse. Likewise, a higher share does not demonstrate more website visits or better enquiries.
Use share as a reason to ask a focused question. Is the relevant page still current? Has the business changed its offer? Are there other observations that suggest a useful content gap? Keep the answer grounded in evidence you can inspect rather than inventing a story about unnamed competitors taking your customers.
Compare periods without assigning causation too quickly
The Compare feature overlays an earlier period so you can examine movement. Choose periods that make sense for the business and record the selection. A comparison can reveal a pattern worth investigating, but the chart does not isolate the effect of a single page edit or marketing activity.
Maintain a change log for meaningful website updates, service changes and measurement adjustments. If a guide was substantially revised, note what changed and why. If several pages changed together, do not attribute the whole movement to whichever edit sounds most impressive in a report.
Demand can also change while your website stays the same. A subject may become more relevant to users at a particular time, or the mix of questions may shift. Treat those as possible explanations to assess, not convenient excuses or proven causes. The reporting should be clear about what is observed and what remains uncertain.
For small counts, show the absolute change alongside any percentage. A movement from one citation to three is worth recording, but it is still two additional observed citations in the report. The scale should influence the response. A modest signal may justify monitoring or a quick accuracy check rather than a large content project.
Investigate a cited page from the reader’s perspective
Open the cited page and examine what a reader would find. Is the main answer current? Are professional claims supported? Does the page explain the scope of the service or product accurately? If it contains a comparison, are the comparison criteria clear? Citation activity is a useful reason to review a page, not proof that every part of it is trustworthy.
Look at the next step too. A technical article may need a relevant route into a product enquiry, while a service page may need clearer coverage or booking information. The improvement should fit the reader’s likely task. Adding a sales form in the middle of every explanation can make a useful page less usable.
Our guide to Bing and Copilot visibility foundations covers the broader website work. When a genuine update is made, IndexNow can notify participating search engines of changed URLs. That notification does not guarantee indexing or a future citation.
Keep the action record specific. “Updated discontinued product references and clarified the enquiry route” is more useful than “Optimised for AI”. It tells the owner what improved for readers and gives the team something concrete to review later, even if citation activity remains variable.
When examining relationships between pages and grounding queries, keep the direction of the filter visible. The current help material describes selecting a query to inspect associated pages or selecting a page to inspect associated queries. Filtered exports reflect that selection. A file saved without its filter context can therefore be mistaken for the whole site’s activity.
Do not force every filtered view into a perfectly reconciled ledger. Bing explains that sampling can produce different counts depending on how the relationship is viewed. The useful question is whether the evidence supports a consistent directional observation, not whether an analyst can invent an adjustment that makes every number match. Preserve the reported values and explain the limitation.
A practical working record can use one row per investigation: the selected page or query, the date range, the observation and the proposed review. If the same subject appears through several views, link those investigations rather than adding their totals together. This prevents the same underlying activity from being counted repeatedly in an owner’s summary.
Also separate an absent row from a confirmed absence of activity. A page that does not appear in a selected view may have limited activity, fall outside the filters or be affected by the report’s coverage. That finding can justify checking the page and the selection, but it cannot establish a penalty by itself. Clear labels keep a missing row from becoming an unnecessarily alarming diagnosis.
Work through a complete hypothetical report review
Imagine a hypothetical business selling workplace storage systems. Its team notices that an older guide to planning storage space appears in the report, while a newly published product page receives little visible citation activity. The numbers are not the starting point for a ranking contest. The team first asks what each page is designed to help a reader do.
The guide explains measurements and planning questions, which makes it useful during early research. The product page contains current specifications but little explanation of selection criteria. The team identifies an opportunity to connect the two more clearly and improve the product information where customers genuinely need clarification. It does not copy every topic label into a new heading.
Next, the team checks whether the older guide contains obsolete product references. It corrects those and records the publication date of the change. It also reviews the enquiry form so a prospective customer can share the information needed for a useful conversation without submitting unnecessary personal details.
At the next review, the business compares the same reporting scope and separately examines website enquiries. A rise in citations would be recorded as visibility movement. Better qualified enquiries would be recorded as a business outcome. The team can discuss whether the pattern is encouraging without claiming that the report proves the content edit caused every enquiry.
Keep citations separate from visits and commercial outcomes
Bing’s citation reporting is not a click or engagement report. Do not multiply citation counts by an assumed click rate and present the result as measured traffic. Unless the rate is part of a clearly labelled scenario, that calculation introduces a number the platform did not provide.
Website analytics and customer records can add evidence, but their boundaries still matter. A recorded session from an assistant is different from a citation shown in an answer. An enquiry may occur later through another route. The business can value those observations without forcing them into a complete customer journey that the available data cannot establish.
The article on AI mentions and business value explains how to handle that commercial interpretation. A useful report might show citation trends, relevant landing page activity and qualified enquiries in separate sections, with the relationship described cautiously.
If the owner asks whether the work is worth continuing, respond with the decisions and evidence available. Are important pages more useful? Are factual errors being corrected? Are observed visits and enquiries relevant? Those questions are more useful than assigning a monetary value to each citation without a defensible basis.
Make the next review proportionate to the evidence
A sensible reporting routine gives attention to consequential changes and avoids reacting to every fluctuation. Review the important pages, note the topics that align with your business and investigate material accuracy problems. Leave a useful, current page alone when there is no clear improvement to make.
Ask your analyst to show the evidence behind the interpretation. Which filters were used? Is the feature still in preview? Are counts being compared at the same level? Is a grouped query being mistaken for an exact prompt? Can the proposed action be justified by a real reader need?
The wider framework for measuring AI visibility brings these questions together without inventing a universal rank. Bing AI Performance provides valuable information within its defined scope, and that information is most useful when the business keeps its decisions equally well defined.
Star Digital AI Marketing offers AI visibility reporting that connects platform observations with practical content priorities. Contact our team if you want help reading the report, checking the pages involved and deciding what deserves attention in your next review.
