AI Search

How to Track Bing and Copilot Citations

5 September 2026 10 min read
Short answer

Bing Webmaster Tools now includes an AI Performance report that shows when pages from a website are cited across Microsoft Copilot, AI-generated summaries in Bing and selected partner experiences. It reports citation activity, cited pages, grounding queries, topics, intents and citation share. This provides genuine evidence of AI visibility, but it does not turn citations into rankings or prove that they generated traffic.

How Do You Track Bing and Copilot Citations Reliably?

Bing Webmaster Tools can now show when pages from your website are cited across Microsoft Copilot, AI-generated summaries in Bing and selected partner experiences. The AI Performance report includes citation activity, cited pages and grounding queries, with preview features extending analysis into intents, topics, citation share and comparisons. Microsoft now offers real evidence of AI visibility rather than guesswork.

This provides evidence that was previously dependent on screenshots and third-party tracking. It does not turn citations into rankings or prove that they generated traffic.

Key takeaways

  • Bing's AI Performance report (public preview, February 2026) shows citation activity across Copilot, Bing AI summaries and partner experiences.
  • Total citations is a visibility count, not a position, prominence, authority or traffic metric.
  • Average cited pages shows breadth of coverage, not placement within any single answer.
  • Grounding queries reveal language and subtopics that ordinary keyword tracking misses, but are a sample, not an exhaustive list.
  • Intents, Topics, Citation Share and Compare add analytical lenses but should not be mistaken for universal share of voice.
  • Build a layered dashboard: executive outcome, AI visibility, opportunity and technical confidence.
  • Establish a dated baseline export before making changes, and use at least four weeks where possible.
  • Never claim causation from a citation increase after an edit without a controlled comparison.

What does the AI Performance report measure?

Microsoft introduced AI Performance in public preview in February 2026. The report focuses on how website content participates as a source in supported AI-generated answers.

Total citations

Total citations show how often URLs from the site appeared as sources during the selected period.

This is a visibility count. It does not indicate:

  • citation position;
  • the prominence of the link;
  • authority;
  • the importance of the supported claim;
  • a unique user count; or
  • visits and conversions.

If one URL is cited several times, those appearances can contribute repeatedly. Always retain the denominator and date range when comparing periods.

Average cited pages

This represents the average number of unique pages from the website displayed as sources per day during the selected period.

It helps distinguish a site whose visibility depends on one article from a site with broader citation coverage. It still does not show the role or placement of each page within an individual answer.

Grounding queries

Grounding queries are phrases used by the AI retrieval process when locating content referenced in generated answers. Microsoft describes this as a sample of overall citation activity.

They can reveal language and subtopics that ordinary keyword tracking misses. A page written for "business succession planning" might be cited through grounding phrases about valuing shares, funding a buyout or protecting a company after an owner's death.

Treat these phrases as research evidence, not an exhaustive keyword list.

Page-level citation activity

The report shows how frequently individual URLs are cited. Use this to identify:

  • consistently useful source pages;
  • unexpected pages representing an entity;
  • high-value pages receiving no citations;
  • outdated URLs still being selected; and
  • visibility concentrated in a fragile part of the site.

Visibility trends

The timeline shows how citation activity changes. Annotate content updates, migrations, indexing incidents and known platform changes before interpreting movement.

What do Intents, Topics, Citation Share and Compare add?

Bing's expanded preview reporting provides four additional analytical lenses. Understanding them properly matters just as much as knowing how AI Overview optimisation works on Google, since both platforms reward publishers who track evidence rather than assumptions.

Intents

Intents group grounding activity by the type of need behind it, such as learning, researching, solving, navigating, buying or finding local information.

The labels should guide interpretation rather than dictate content. If commercial citations grow while informational citations fall, the overall count may hide a meaningful improvement.

Topics

Topics group related citation activity into themes. This helps marketers examine subject coverage without treating every phrase as an independent keyword.

Compare topic visibility with the site's strategic content clusters. A financial services website might separate retirement, protection, investment and business-owner topics, then assess which cluster is actually earning citations.

Citation Share

Citation Share indicates how much of the observed citation opportunity the site captures within the report's scope.

State the scope whenever reporting it. It is not universal share of voice across every AI engine, region or conversation.

Compare

Comparison tools allow periods, segments or performance groups to be examined side by side. Use equal date ranges and account for seasonality.

How should Bing AI visibility be reported?

Build a layered dashboard.

A layered Bing AI visibility dashboard
LayerWhat it covers
1. Executive outcomeQualified visits, enquiries, conversions, assisted outcomes, revenue where defensible, branded demand changes
2. AI visibilityTotal citations, average cited pages, citation share, priority topics, intent distribution, most cited URLs, meaningful trends
3. OpportunityUncaptured valuable topics, partly answered grounding queries, uncited commercial pages, evidence gaps, outdated URLs, fragile clusters
4. Technical confidenceIndex coverage, crawl issues, canonical conflicts, sitemap health, IndexNow implementation, duplicate content, release annotations

1. Executive outcome

Show:

  • qualified visits from relevant Microsoft surfaces where identifiable;
  • enquiries or conversions;
  • assisted outcomes;
  • revenue where attribution is defensible; and
  • notable changes in branded demand.

Do not imply that every conversion after a citation was caused by it.

2. AI visibility

Report:

  • total citations;
  • average cited pages;
  • citation share;
  • priority topics;
  • intent distribution;
  • the most cited URLs; and
  • meaningful trends.

3. Opportunity

Identify:

  • valuable topics where competitors or other sources dominate;
  • grounding queries the site partly answers;
  • commercially important pages without citations;
  • evidence gaps;
  • outdated cited URLs; and
  • clusters relying on one page.

4. Technical confidence

Include:

  • index coverage;
  • crawl issues;
  • canonical conflicts;
  • sitemap health;
  • IndexNow implementation;
  • duplicate content; and
  • major release annotations.

This prevents a content team from rewriting pages when the real problem is discovery or canonicalisation.

Establish a reliable baseline

Before making changes, export the current report and record:

  • reporting period;
  • active filters;
  • supported surfaces described by Microsoft;
  • total citations;
  • average cited pages;
  • top pages;
  • priority grounding queries;
  • topic and intent mix;
  • citation share; and
  • known limitations.

Keep the raw export. A dashboard interface can change, but a dated baseline allows a fair before-and-after comparison.

For a new programme, use at least four weeks where possible. A shorter window can be directionally useful but should not be presented as a stable benchmark.

Segment by business value

Not every citation deserves equal weight. Tag pages and queries by:

  • funnel stage;
  • service or product;
  • customer type;
  • market or location;
  • informational versus commercial intent;
  • page owner;
  • freshness sensitivity; and
  • conversion potential.

Suppose a consultancy receives 1,000 citations for a glossary page and 40 citations for a guide used during supplier selection. The glossary wins by volume. The supplier guide may be more valuable commercially.

Use a weighted opportunity score if stakeholders need prioritisation, but keep the underlying counts visible.

Connect citations to analytics carefully

AI exposure often occurs without a click. Standard analytics can measure sessions that arrive at the website, not everyone who saw a citation.

Track:

  • identifiable referral traffic;
  • landing pages;
  • engagement;
  • conversion rate;
  • assisted conversions;
  • branded organic searches;
  • direct traffic trends; and
  • enquiry-source responses.

Avoid claiming that an increase in direct traffic was caused by Copilot without supporting evidence. It may be consistent with greater awareness, but it remains an inference.

Microsoft has argued that AI-search journeys can be shorter and more conversion-oriented. Treat platform-level claims as context rather than a forecast for every website.

Run a controlled content test

Choose a small group of pages and document:

  • the intended grounding queries;
  • current citations and cited-page data;
  • technical eligibility;
  • the exact content changes;
  • the submission or IndexNow date;
  • observation windows; and
  • downstream engagement.

Improve substance, not merely formatting. A valid test might add a current dataset, clarify a decision framework and cite primary evidence. Changing three headings and adding an FAQ is unlikely to establish why performance moved.

Use a comparison group of similar unchanged pages where possible. Search systems, demand and competing sources change over time, so a simple before-and-after chart cannot prove causation. The same discipline applies to structured AI Overview optimisation work on Google, where controlled testing is equally important.

How often should data be reviewed?

Use different cadences for different jobs:

  • Weekly: technical incidents, major launches and high-priority volatile topics.
  • Monthly: citations, topic coverage, grounding queries and business outcomes.
  • Quarterly: content investment, cluster resilience and strategic opportunity.

Constant manual checking encourages reactive edits. A stable page can rotate in and out of generated answers without anything being wrong.

Common reporting mistakes

Calling citations rankings

A citation count does not reveal a numbered position. Use Microsoft's terminology accurately.

Ignoring the denominator

Fifty citations could represent most available opportunities or a tiny fraction. Use Citation Share and the size of the tracked topic set where available.

Combining unlike periods

Do not compare 28 days with a 31-day period without normalising. Consider seasonality and changes in demand.

Treating every citation as commercial

Separate informational visibility from purchase or enquiry-related intent.

Claiming causation

A citation increase after an edit is evidence of sequence, not proof that the edit caused the increase.

Treating Bing data as the entire AI market

Bing Webmaster Tools reports supported Microsoft and partner experiences. It does not represent Google, ChatGPT, Perplexity or every AI answer on the web.

Sources

Frequently asked questions

Is AI Performance available in Bing Webmaster Tools?+

Yes. Microsoft launched the report in public preview in February 2026 and has continued expanding its analytical capabilities.

Does Total Citations show clicks?+

No. It counts displayed source citations within the report's supported scope. Use analytics and conversion systems for visits and outcomes.

What is a grounding query?+

It is a phrase used in retrieving information that supported an AI-generated answer. Microsoft notes that the report shows a sample rather than every retrieval event.

Is Citation Share the same as organic share of voice?+

No. It concerns citations within the defined AI-reporting scope. Conventional rank tracking uses different observations and should be reported separately.

Can the report prove a content update worked?+

Not by itself. Use documented changes, comparable periods, control pages and several metrics. Even then, describe the strength and limitations of the inference.

How does this compare with tracking Google AI Overviews?+

The principles are similar: track visibility, avoid conflating citations with rankings, and connect evidence to analytics carefully. See our guide on how to track AI Overview visibility for the Google-side approach.

How often should I check the AI Performance report?+

Weekly for technical incidents and major launches, monthly for citations and topic coverage, and quarterly for strategic content investment decisions.

Does a rise in citations always mean more revenue?+

No. Citations indicate visibility within supported Microsoft experiences, not clicks or purchases. Revenue should be tracked separately through analytics and enquiry-source data, then related to citation trends rather than assumed to be caused by them.

Measure visibility without overstating it

Bing's AI Performance reporting gives publishers a valuable view of how their content contributes to generated answers. Its strength is not one headline number. It is the ability to connect cited pages, grounding language, topics, intents and citation share.

Use that evidence to prioritise improvements and ask better commercial questions. Keep citations, visits and conversions separate, then explain how they relate. If you'd like help building a dashboard that connects Bing citations, Google AI Overview visibility and genuine commercial outcomes, talk to us.

Related reading

Turn AI search visibility into measurable pipeline.

A short review shows where your site is already close to being cited, and what to fix first.