Bing AI Performance: The Complete Guide to Tracking Microsoft Copilot Citations, Share of Voice, and AI Search Opportunities

Microsoft gives you a metric for your share of citations in AI answers. It’s called Citation Share, and you’ll find it in Bing Webmaster Tools. The useful part is how clearly Microsoft explains what it measures and what it leaves out.

It’s also a preview feature. Its limits are clear, and several contradict the advice you may have read elsewhere.

Let’s separate Citation Share from Total Citations first. They come from different releases and measure different things. Then we’ll look at Topics Portfolio’s opportunity queries, what IndexNow can do, and where Copilot fits into shopping. The operating model will help you work out which part of the system needs attention.

The three releases that explain the confusing metric names

Most confusion about Bing AI reporting comes from a naming collision across three separate launches. If you have read conflicting descriptions of these metrics online, the conflict is almost always this: two different metrics, two different release dates, one report.

Release one - AI Performance, public preview, 10 February 2026. Microsoft’s announcement post describes a dashboard answering “where is my content being cited in AI-generated answers?” It defines five things: Total Citations, Average Cited Pages, Grounding queries, page-level citation activity, and visibility trends over time. None of these is a share metric.

Release two - Intents, Topics, Citation Share, Compare, preview, June 2026. A separate post, New AI Visibility Insights in Bing Webmaster Tools, added four capabilities that Microsoft describes as available “in preview globally.” Citation Share arrived here. So did Intents (grounding queries classified by intent), Topics (related queries grouped into themes), and Compare (overlay a previous period on the current chart).

Release three - Topics Portfolio, beta, 13 October 2026. The Topics Portfolio announcement adds a beta tab carrying topic-level ecosystem measures and Opportunity Queries - queries where other sites appear and yours doesn’t.

So the precise answer to the naming question: Total Citations and Citation Share are two different metrics in the same report. Total Citations is a volume count aggregated across supported AI surfaces over a date range. Citation Share is a per-query ratio - your citations divided by all citations shown for one specific grounding query. Microsoft’s own help documentation puts it in a two-row table:

Metric	What it tells you
Total Citations	Volume - how often your content is referenced overall
Citation Share	Relative citation presence - your share of citations for a specific grounding query

That same help page answers the follow-up directly: “A site may have a high total citation count but a lower Citation Share when citation activity for the same query is distributed across many sources.”

One clarification on a widely repeated line. The June post does say Citation Share “does not expose competitor domains” - the full sentence reads “It does not expose competitor domains, represent traffic share, or assign quality scores to content.” The official help page states the same restriction more plainly: Citation Share “does not expose competitor domain names or identify which other sites hold the remaining share.” Both statements are accurate and current. Neither is contradicted by the later Topics Portfolio beta, which reports how many sites appear in a topic (a count) without ever naming them.

What AI Performance measures across which surfaces

The help documentation is unusually precise about scope. AI Performance “reflects grounding and citation activity across the following supported AI surfaces”:

• Microsoft Copilot
• AI-generated summaries in Bing
• Select partner AI integrations

That third bullet is worth pausing on. Microsoft names the first two surfaces explicitly and leaves the third generic. The help page doesn’t enumerate which partner integrations are included.

The metrics, defined on the help page:

Total Citations - “The total number of times your content was visibly referenced or shown as a source in AI-generated answers during the selected date range.”

Cited pages - “The number of unique pages from your site that were cited on a given day.”

Average cited pages - “The average number of unique pages that were cited per day over the selected time range.”

Grounding queries - “The key phrases the AI used when retrieving content that was cited in its answer.”

Page-level citation activity - citation counts by specific URL, which “reflect how often pages are cited, not their importance, ranking, or role within a response.”

Available date ranges are 7 days, 30 days, 3 months, or a custom range within available history. Data is “refreshed daily, with a short processing delay.”

The explicit non-claims: what this report will never tell you

This is the section most worth reading, because the report’s limitations are unusually well documented and rarely reported accurately. Microsoft has published a set of flat denials.

The help page states, in a highlighted note: “AI Performance does not measure rankings, authority, performance, or importance. It simply shows which content was cited.”

On trend interpretation, another note reads: “These trends reflect aggregated citation activity only and cannot be attributed to a specific cause or event. Changes in citation trends are observational and do not indicate the impact of any single update, model change, or content modification.”

On traffic, from the FAQ: “Do citations represent clicks or traffic?” - “No. A citation indicates that your content was visibly referenced or shown in an AI-generated answer. It does not represent traffic, clicks, or user engagement.”

On traditional metrics: “Do AI Performance metrics match traditional search metrics?” - “No. Grounding and citations show how AI systems use your content to generate answers. This differs from traditional search metrics, which measure rankings, clicks, and traffic.”

On competitor identity: Citation Share “does not expose competitor domain names or identify which other sites hold the remaining share.”

On causation, for both Compare and content changes: Compare “shows what changed between two periods, but it does not explain why the change happened.” Citation Share “can help you observe whether your share changes over time after content updates, but it does not establish a causal link.”

The sampling caveat that changes how you read an empty dashboard

This is the practical trap. The help page states plainly:

AI Performance data is aggregated and summarized to provide a representative view of citation activity across supported AI experiences. The data shown is not a complete log of every instance where content may have been referenced.

And then lists the consequences:

• Not all citation activity may appear in the dashboard.
• Very low or infrequent citation activity may not surface.
• Totals may differ across views (for example, between pages, grounding queries, and time-series charts).
• The data shown represents a sample of overall citation activity.

There’s a further, sharper problem when you cross-reference. If you filter by a grounding query and then look at a page, you may see a different citation count than if you filter by that page and view the same grounding query. The explanation: “AI Performance data is sampled, and grounding queries and pages may each be sampled over slightly different time windows within your selected date range.”

And on empty results: if you see no grounding queries, “very sparse or infrequent citations may not surface in the dashboard. This is expected and does not indicate a penalty or exclusion.”

A blank AI Performance dashboard doesn’t tell you whether your site is indexed, excluded, or penalised. You might have too little citation activity to meet the sampling threshold. You might not be indexed at all. The report can’t tell those situations apart because it only reports content that made it through.

Grounding queries are grouped phrases, not user prompts

This trips up anyone arriving from classic keyword rank-tracking, and the help page is blunt about it. Grounding queries “are not full user questions or prompts.” They’re “grouped, generalized phrases used to summarize citation activity across AI-generated answers,” and the tool “does not display individual user questions or highly specific long-tail queries.”

Three structural facts follow, all documented:

• A single grounding query can be associated with multiple pages from your site.
• A single page can appear under multiple grounding queries if it was referenced in different answers.
• Grounding queries “may be determined differently across supported AI experiences and partners, and the phrasing shown may vary.”

The report includes a Grounding Query–Page Mapping view connecting the two: select a grounding query to see its cited pages, or select a page to see its associated grounding queries. Two constraints apply. You can filter by one or the other, not both simultaneously - “Export the filtered data for further cross-referencing.” And exports reflect whatever filter is currently applied.

Intents and Topics sit one level above

Intents classify grounding queries by intent using AI/ML classifiers. The documented taxonomy goes well past the classic informational/navigational/transactional split. The help page lists: Informational, Media, Navigational, Commercial, Learn and Solve, Research, Live Event, Local, Comparison, Planning, Utility, Creation, Conversational, and Other.

Microsoft’s own examples of use are worth reading as intent-positioning advice: “If your site is monetized through e-commerce, strong citation activity for Commercial or Comparison intents can help you understand whether your content is appearing in commercially oriented AI experiences.” And for gaps: “If you expect to see citation activity for Research queries but mostly see Informational ones, that may indicate an opportunity to add more depth, supporting evidence, or clearer coverage.”

The docs also recommend matching format to intent - “comparison tables for Comparison, step-by-step guidance for Planning, or concise explanations for Informational queries” - while cautioning that these insights “do not guarantee citation outcomes.”

Topics group related grounding queries into thematic clusters. Microsoft defines the hierarchy plainly: “Topics sit one level above them by grouping related grounding queries under a common theme.” Microsoft warns the classifiers are imperfect, “especially for niche or highly specialized domains,” and that filtering a topic to see cited pages requires exporting and filtering externally.

Both Intents and Topics are classifier-driven, which means both inherit classifier error. Microsoft’s own guidance is to use thumbs up/down feedback inside the view, which it says “go straight to the team working on the classification.”

Reading Citation Share without fooling yourself

Citation Share is the most valuable thing Microsoft has shipped here and the easiest to misread. Microsoft’s help page publishes an interpretation guide, which is the best available starting point:

• High and stable share - your site is receiving a consistent share of citations for that grounding query.
• High but declining share - citation activity may be becoming more distributed across sources.
• Low share on a core query - you’re receiving a smaller portion of the citation space for a query that matters to you.
• Growing share - you’re receiving a larger share over time, “which may reflect changes in content, user demand, the broader web ecosystem, or AI system behavior.”

The doc’s own warning is the important part: “Changes in share can be driven by many factors, including shifts in user demand, content changes across the web, model updates, freshness signals, and partner refresh cycles.”

The June announcement frames the same limitation from the other direction, calling Citation Share “an observational metric – not a ranking system or a competitive scoreboard.”

One consequence deserves emphasis for anyone reporting on this tool: because Citation Share is a ratio whose denominator is “all citations shown across all sites for that same grounding query,” a decline in your share doesn’t necessarily mean competitors improved. The denominator can move. Microsoft lists “partner refresh cycles” and “broader changes across the web itself” among the drivers. Treating a share drop as a competitive loss is an inference the data doesn’t support.

Topics Portfolio and the Opportunity Queries beta

The newest layer, announced 13 October 2026, is the first part of AI Performance that gestures at competitive whitespace. The post is candid about its own maturity, and that candour is worth preserving in any summary of it.

The landing page lists a personalised set of topics where your site has meaningful presence. Each row carries three ecosystem measures, defined in the post:

Column	What it tells you
Topic	A broad thematic cluster of related grounding queries where your site appears
Topic Size	The approximate total citations across all sites for this topic over the selected period
Sites Cited	The approximate number of sites appearing for queries in this topic
Queries	The approximate number of distinct grounding queries in this topic

Opportunity Queries are “the strongest queries across your whole portfolio where other sites appear and yours does not.” Inside a topic you get two tables: Grounding Queries (where you already appear) and Opportunity Queries (where you don’t). Microsoft’s framing: “Together they give you the two halves of a topic: what you already own, and what is still open.”

Five limitations Microsoft publishes, reproduced here because they constrain how far you should trust the view:

1. Topic classification is AI-generated and can occasionally be imperfect, “so a query may land in a topic you would have labelled differently.”
2. “Topics with insufficient site presence or insufficient ecosystem activity are not shown, so smaller sites may see a shorter list or may not even see this feature.”
3. “Branded queries can appear as opportunity queries” - Microsoft’s own example has contososhoes.com seeing “Air Shoes” listed as an opportunity inside a Running Shoes topic even though it’s a branded competitor query.
4. Topic Size, Sites Cited and Queries are approximate.
5. Opportunity citation volumes are “deliberately shown as ranges.”

Microsoft’s own instruction on how to treat the output: “Treat the numbers as directional signals for planning, not as a precise ledger.”

If you run a small site, a missing Topics Portfolio tab doesn’t tell you how visible you are. You may not meet its threshold. The main report also samples its data. That’s two reasons a dashboard can be blank, so check both before drawing a conclusion.

The Bing/Copilot visibility operating model

Microsoft’s October 2025 content guidance supplies the crucial mechanistic detail. It describes the process as parsing: “Assistants like Copilot break content down, a process called parsing, into smaller, structured pieces that can be evaluated for authority and relevance. Those pieces are then assembled into answers, often drawing from multiple sources.” The same post draws the distinction that matters: “In traditional search, visibility meant appearing in a ranked list of links. In AI search, ranking still happens, but it’s less about ordering entire pages and more about which pieces of content earn a place in the final answer.”

That gives six layers. The right-hand column shows what each layer costs you when it fails - the first three are invisible in AI Performance.

Layer	Question it answers	Your controls	What fails when it breaks	Visible in AI Performance?
0 - Admission	Is the URL in Bing’s index at all?	Crawlability, robots.txt, XML sitemap, IndexNow	Total absence from every AI surface	No
1 - Legibility	Can the page be broken into assessable pieces?	Headings, self-contained phrasing, tables, Q&A blocks, JSON-LD schema, HTML over PDF, no content hidden behind tabs	Indexed, retrieved sometimes, never assembled into an answer	No
2 - Retrieval	Does the content get fetched for the right phrases?	Topical coverage, entity consistency across text/image/video, freshness	Retrieved for adjacent phrases only, or rarely	Partially - via Grounding queries and Topics
3 - Selection	Does it earn a place in the assembled answer?	Depth, specificity, evidence, current information	Retrieved but passed over for a competitor’s page	Yes - Total Citations, Citation Share
4 - Ecosystem	Who else is there, and where is the gap?	Content gap analysis	No strategy input at all	Yes - Topic Size, Sites Cited, Opportunity Queries (beta)
5 - Commerce	Does it sell inside AI surfaces?	MMC product feed, UCP attributes, Copilot Checkout eligibility	Citable but not purchasable	No - use MMC, not BWT

Three things fall out of this model that I think are genuinely useful and aren’t stated anywhere in the Microsoft documentation.

AI Performance only measures layers 3 and 4. It’s a selection-and-ecosystem instrument, not a visibility instrument. When your dashboard is empty, the cause is far more likely in layers 0, 1 or 2 - which the report structurally can’t see. Anyone treating “no citations” as a content-quality verdict is reading a layer-3 instrument and generalising to layers 0–2.

Layers 0–2 fail silently; layers 3–4 fail visibly. A page blocked in robots.txt generates no citation data and no error. A page that’s indexed, parseable, retrievable but out-competed shows up as a low Citation Share. Both look like “low visibility” in a dashboard, and they call for opposite responses: fix admission or structure versus improve competitive depth.

The two clocks don’t align. IndexNow operates on a crawl-latency clock measured in seconds to minutes. AI Performance refreshes daily with a processing delay and reports sampled aggregates. You therefore can’t use AI Performance to evaluate whether an IndexNow submission worked. Microsoft’s Compare documentation says as much in its own terms: “Compare helps publishers observe patterns, not establish causation.” Practically: judge IndexNow on IndexNow and crawl-coverage evidence, and judge AI visibility on citation trends, with a lag of weeks between any content change and a visible citation shift.

IndexNow: the fastest lever, and exactly what it cannot do

IndexNow is the indexing-signal layer under the whole Microsoft stack, and Microsoft’s AI Performance launch post explicitly recommends it, framing it as keeping content fresh “across search and AI experiences.”

The protocol documentation is the authoritative spec. A single-URL submission is an HTTP GET to a participating engine:

https://<searchengine>/indexnow?url=url-changed&key=your-key

Keys must be 8–128 characters. The documentation describes them as hexadecimal, though the character set it then permits - letters a-z and A-Z, digits and dashes - is wider than strict hex, and the FAQ states the length rule without the hexadecimal label. Either way, the practical constraint is the charset: alphanumeric plus hyphens, in a UTF-8 file named after the key. Batch submission is a POST with JSON, capped at 10,000 URLs per request. Ownership is proven by hosting a UTF-8 key file, ideally at the host root; the keyLocation parameter exists for non-root placement, but the docs note the constraint it creates - “a key file located at http://example.com/catalog/key12457EDd.txt can include any URLs starting with http://example.com/catalog/ but cannot include URLs starting with http://example.com/help/” - and recommend Option 1, root placement.

Response codes are defined: 200 OK, 202 Accepted (key validation pending), 400 Bad Request, 403 Forbidden (key not valid), 422 Unprocessable Entity (URLs don’t belong to host or key doesn’t match schema), 429 Too Many Requests.

What IndexNow does not do

The FAQ is more candid than the spec, and its limits are the part worth internalising.

It doesn’t guarantee indexing. “Submitting a URL does not guarantee immediate indexing. However, IndexNow increases the likelihood that important changes are discovered and crawled faster.” Reasons a URL may still not be indexed include crawl quota limits, low-quality or duplicate content, technical issues like 404s, robots.txt blocks or noindex tags, crawl scheduling delays, and the simple fact that “the search engine may choose not to index a URL based on its algorithms.”

HTTP 200 means almost nothing on its own. The documentation is explicit: “The HTTP 200 response code only indicates that the search engine has received your URL.”

It consumes crawl quota. “Every URL submitted through IndexNow counts toward your site’s crawl quota.” The upside is that submission is itself a prioritisation signal, but the budget is real.

It isn’t a sitemap substitute. “IndexNow is intended for notifying search engines about recently added, updated, or deleted URLs. It is not designed for submitting every URL on your site at once.” The documented split: IndexNow for high-priority or frequently changing pages, sitemaps for complete inventory. The docs recommend both together, and recommend against retroactive submission of changes predating setup.

It wants debouncing. Wait at least 5 minutes between resubmissions of the same URL; after a 429, wait at least 10 minutes. Repeatedly submitting unchanged URLs “may lead to wasted crawl quota without improving visibility.”

Rate limits aren’t public. “IndexNow does not publicly disclose exact rate limits, as each participating search engine sets its own daily submission thresholds per site.”

Google does not participate

Check the participating endpoints before you build around IndexNow. The FAQ lists them: the IndexNow global endpoint (api.indexnow.org), Amazon, Bing, Naver, Seznam.cz, Yandex, and Yep.

Google isn’t among them. The docs also note that “submitted URLs will be automatically shared with all other participating search engines” - so submitting once reaches the whole IndexNow network, none of which includes Google. If your priority traffic is Google, IndexNow isn’t a Google optimisation. It’s a Bing-and-partners optimisation, and one whose value shows up in AI surfaces built on Bing’s index.

Native CMS support is broad - the FAQ names WordPress, Shopify, Wix, GoDaddy, Joomla, Drupal, PrestaShop, Shopware and others, with Cloudflare offering native CDN integration - so for most sites this is a configuration task rather than an engineering one.

Copilot as a shopping and search surface

Copilot isn’t only an answer engine. It’s a transaction surface, and the reporting you have read so far covers the citation half, not the commercial half.

Copilot Checkout: what is actually live

Microsoft’s January 2026 announcement introduced Copilot Checkout and Brand Agents. The announcement states Copilot Checkout was “beginning to roll out in the U.S. on Copilot.com, with partner activation across PayPal, Shopify, and Stripe.” Shopify merchants, per the post, “will automatically be able to sell in Copilot Checkout. No application. No integration.” The merchant remains merchant of record.

The announcement also contains forward-looking claims about reach that the current product page doesn’t support. The January post says “As Copilot Checkout expands across the Copilot ecosystem (Bing, MSN, Edge, and more), you’re automatically there.” The live Agentic Commerce page answers the question “What surfaces is Copilot Checkout live on?” with: “Copilot Checkout is available on Copilot.com and the Copilot mobile app, with plans to expand to additional surfaces soon.”

That changes the picture. The current page doesn’t list Bing, MSN or Edge as live Checkout destinations. If your traffic forecast includes all of them, the January announcement makes the available reach look larger than it is today.

Eligibility on the current page is narrow and worth stating plainly: “Only English-language merchants who sell to US buyers are eligible at this time (supporting USD).”

The product feed layer, and the honest status of UCP

Microsoft’s stated mechanism for organic product visibility in Copilot appears in the Agentic Commerce FAQ, under the question “What information do you use to surface my product in your organic results?” The answer is the sentence that should shape feed strategy:

We leverage both information found on the web and from a merchant’s feed in the Microsoft Merchant Center.

This is the bridge between the content half and the commerce half of this guide. Organic Copilot product discovery draws on both your crawlable web content and your Microsoft Merchant Center feed. Neither substitutes for the other.

The same page describes the setup path: add store settings in MMC (return policy, customer support, Copilot Checkout), then “Update your feed with attributes needed for UCP - these additional feed attributes include: native checkout eligibility for products, product warnings if needed, and Merchant Item ID (unique ID).”

On Universal Commerce Protocol, the same page says MMC “will support Universal Commerce Protocol (UCP), enabling richer signals (returns/support policies) so AI can assess products with confidence.” The verb is future tense. Treat UCP support in MMC as announced and directional, not as a generally available capability you can build against today.

One more status marker on that page. In the numbered setup steps, “Copilot Checkout” appears as a store setting followed by an asterisk, and the footnotes resolve it: ”[*] Pilot only to select customers.” So the MMC-side checkout control is a pilot, distinct from the broader Checkout rollout described in the January post. Announced, pilot, and generally live are three different states, and Microsoft marks them differently on the same page.

Microsoft also says, in the January post, that “Although MMC is not required to sell through Copilot Checkout, product feeds from MMC help inform organic Copilot results.” That’s the sentence to quote when someone asks whether they need a merchant feed at all.

How the Bing stack differs from Google and OpenAI

Microsoft’s positioning claim is unusually broad, and worth reading precisely rather than paraphrasing loosely. In Elevating the Role of Grounding on the AI Web, written by Jordi Ribas, Corporate Vice President, Search & AI, the claim is that “Microsoft grounding powers nearly every major AI assistant in the market.”

The Microsoft Web IQ product page makes the same claim in commercial form: “Web IQ powers leading frontier AI assistants, including Microsoft Copilot and ChatGPT.” It describes Web IQ as “a suite of AI-native web search and grounding APIs,” drawing on “Bing’s global web index” and combining “the open web, licensed content, and structured data.”

If that holds, your Bing work could reach beyond Copilot. Bing’s index and Microsoft’s grounding layer also serve other assistants. Making your content easy to understand and cite there may help it appear on some non-Microsoft AI platforms too.

Two constraints keep that honest. First, the AI Performance report doesn’t tell you that ChatGPT citations are included - it lists “select partner AI integrations” without naming them, so you can’t verify from the report which surfaces contribute. Second, Microsoft’s performance claims for Web IQ are self-reported: the page cites “approximately 2.5× faster retrieval processing (164 ms p95 latency)” and “50% fewer tokens at comparable answer quality,” both attributed to “In Microsoft benchmark tests.” Those are vendor benchmarks against unnamed comparators and should be treated as positioning, not proof. Web IQ’s published list pricing is $1 per 1,000 calls for Browse and $7 per 1,000 for Web, News, Images and Video.

Set against the other two stacks:

Against Google. Google’s canonical AI optimisation guidance - a page last updated 10 July 2026 - describes a “Generative AI performance report in Search Console” and states that “a site must be included in Search generative AI features in Search Console to be eligible for display in generative AI features on Google Search.” Google also publishes an explicit debunk - “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search.” Google’s guidance emphasises that generative answers rely on core ranking systems and warns against creating separate content per fan-out query as scaled content abuse. The structural difference: Google gates AI appearance behind an explicit eligibility switch, and describes RAG/grounding as resting on core Search ranking over the Search index. Bing’s Citation Share, by contrast, is a measurement surface with no equivalent documented eligibility toggle.

Against OpenAI. In the OpenAI documentation I reviewed - the crawler and bot-policy pages - there’s no first-party citation or share-of-voice dashboard. What that documentation does publish is crawler policy: OpenAI runs four distinct crawler user-agents, blocking GPTBot doesn’t block search inclusion, and blocking OAI-SearchBot removes a site from search entirely. That’s an access-control surface, not an analytics surface. Microsoft’s is currently the reverse - rich analytics, and no equivalent public per-surface opt-out documented beyond standard robots.txt handling, which the AI Performance report explicitly respects: “Bing respects all content owner preferences expressed through robots.txt and other supported control mechanisms.”

The honest summary: as of 19 October 2026, Microsoft’s first-party AI citation reporting is detailed about what it exposes and gates less than Google’s documented approach. The trade-off - rich measurement, lighter gating - is worth naming explicitly if you have compliance or consent obligations.

The tracking-setup checklist

Foundations - do these before you read a single metric

1. Verify your site in Bing Webmaster Tools and confirm URL Inspection shows the URLs you care about as indexed. This is layer 0, and it’s the only layer you can’t inspect from AI Performance.
2. Publish a valid XML sitemap. IndexNow doesn’t replace it, and the IndexNow FAQ explicitly recommends using both.
3. Deploy IndexNow at the host root with a UTF-8 key file. Prefer CMS-native integration if one exists.
4. Confirm your robots.txt allows the crawling you expect. AI Performance is downstream of this and will show nothing if you’re blocked.

Measurement setup

5. Enable IndexNow if not already running, and log HTTP response codes with timestamps. You’ll need this log when you want to prove a submission landed.
6. Read the AI Performance report over a 30-day window first, not 7 days. The tool’s own guidance to the Topics Portfolio beta is to “choose a one-, two- or three-month window.”
7. Establish your baseline: record Total Citations, Average Cited Pages, and the page-level citation table before making any change. Without a baseline, Compare has nothing to compare against.
8. Export to CSV or Excel before you filter. The help page confirms exports exist and that “Exports reflect the currently applied filter” - so filtering then exporting gives you a filtered slice, not the full set.

Interpretation discipline

9. Read Total Citations and Citation Share as different numbers answering different questions. Volume isn’t share. A site can be cited constantly and still hold a thin share of any single query.
10. Treat empty data as uninformative, not negative. Both sparse citation activity and non-indexing produce a blank dashboard, and the report can’t distinguish them.
11. Don’t infer competitor identity from Citation Share. The data doesn’t contain it.
12. Don’t infer causation from Compare. It shows change over time; it doesn’t explain it.
13. Expect totals to disagree between views. Sampling explains this, and Microsoft says so explicitly.

Competitive and commercial layers

14. If Topics Portfolio appears for your site, use Opportunity Queries as a planning input and treat the volumes as ranges. Microsoft says directional, not a ledger.
15. Filter branded-query artefacts out of your opportunity list before acting on it. Microsoft warns these appear by design.
16. If you sell products, audit your MMC product feed independently of content work. Organic Copilot discovery draws on both the web and the feed.
17. Confirm current Checkout eligibility against your own geography and language before investing. As documented, it’s US buyers, English-language, USD.

Where this stack still leaves you blind

Three gaps are worth stating plainly rather than papering over.

The partner surface is unnamed. AI Performance covers “select partner AI integrations” without listing them. You can’t audit what you can’t see.

Layer 2 retrieval is inferred, never measured. Grounding queries tell you which grouped phrases were associated with your citations. They don’t tell you the queries where you were retrieved and passed over, which is the most actionable blind spot in the whole report.

Nothing here measures commercial outcomes. Citation Share is explicitly not traffic, not traffic share, and not a quality score. If you need conversion signal from AI surfaces, this tool is the wrong instrument by Microsoft’s own design - you’d have to source it from your own analytics and from the merchant tooling.

The honest framing for anyone building a measurement practice on this: Bing AI Performance is the most explicit first-party instrument for AI citation visibility I was able to verify, and it’s simultaneously a sampled, aggregated, partly-classifier-dependent view of two layers out of five. Use it for what it can see. Diagnose the rest with traditional index and crawl tooling, where the layer-0 and layer-2 signals still live.

Sources and References

• Introducing AI Performance in Bing Webmaster Tools Public Preview - Microsoft Bing, 10 February 2026
• AI Performance in Bing Webmaster Tools (help documentation) - Microsoft Bing Webmaster Tools
• New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare - Microsoft Bing, June 2026
• Introducing Topics Portfolio (Beta) in AI Performance - Microsoft Bing, 13 October 2026
• Elevating the Role of Grounding on the AI Web - Jordi Ribas, Corporate Vice President, Search & AI, Microsoft
• Optimizing Your Content for Inclusion in AI Search Answers - Microsoft Advertising, October 2025
• Conversations that Convert: Copilot Checkout and Brand Agents - Microsoft Advertising, January 2026
• Agentic Commerce - Microsoft Advertising
• Microsoft Web IQ - Microsoft
• IndexNow Documentation - IndexNow.org
• IndexNow FAQ - IndexNow.org
• Bing Webmaster Tools blog index - Microsoft Bing
• AI optimization guide - Google Search Central, last updated 10 July 2026
• AI features and your website - Google Search Central, last updated 10 December 2025
• OpenAI crawler documentation - OpenAI