Microsoft Clarity now lets you see exactly what search queries triggered AI-powered citations to your website—an unprecedented look into how Bing Copilot grounds its responses in real web content. For years, marketers have guessed at the signals behind organic search rankings. With the AI Citations dashboard, the black box cracks open a little wider: you can finally witness the precise, raw queries that Bing’s language model used to retrieve your page, cite it, and display it to users inside Copilot’s conversational answers.
It’s a pivot moment for SEO. Traditional tools only reveal the keywords that humans type into a search box. But Copilot doesn’t just match keywords; it reasons, plans, and executes multi-step searches through an internal orchestrator. The Clarity AI Citations report surfaces the system’s own underlying queries—called grounding queries—that lead a model to your content. Understanding them means understanding the new anatomy of organic visibility in an AI-first search engine.
What exactly are grounding queries?
In the context of Bing Copilot (and similar retrieval-augmented generation models), a grounding query is the transformation of a user’s natural language question into one or more search-engine-friendly strings. When someone asks Copilot, “What’s the best lightweight laptop for college under $800?” the system doesn’t just feed that exact sentence into a search index. It may break it down into several sub-queries, like “best college laptops 2024 under $800,” “lightest laptops for students reviews,” and “battery life comparison student laptops budget.”
The model then searches Bing using these refined strings, retrieves relevant web pages, and synthesizes an answer with citations. Those intermediary search strings are the grounding queries. They bridge the gap between a conversational prompt and a search corpus that still relies heavily on keyword matching and ranking signals.
Before Clarity lifted the curtain, webmasters had no way of telling which grounding queries mapped to their pages—only the eventual user click (if any). Now, for sites that are cited in Copilot answers, you can see the very grounding queries that brought your page into the model’s crosshairs.
A first look at the AI Citations dashboard
If you already use Microsoft Clarity—the free behavioral analytics tool that records user sessions, heatmaps, and performance metrics—the AI Citations module sits under the “AI” menu in the left navigation (available to projects that have received at least one Copilot citation in the past 30 days). It’s broken into three main panels:
- Citation overview: total citations, unique grounded queries, cited pages, and how these metrics trend over time.
- Grounding queries table: a sortable list of every distinct grounding query that pulled a page from your domain into a Copilot response, along with the number of citations generated, distinct pages cited, and the date of first/last citation.
- Cited pages drill-down: when you click a query, you see exactly which URLs were cited, how many times, and whether users engaged with the citation (clicks).
The dashboard covers data from Bing’s consumer Copilot (copilot.microsoft.com) and the Windows-integrated Copilot experience, though corporate Bing Chat Enterprise citations are not included to protect privacy.
Why this matters for SEO strategy
Traditional SEO revolves around ranking for user-facing keywords. But in an AI-driven world, your page is no longer just a blue link on page one. It becomes a piece of a machine’s evidence chain. Visibility now depends on whether the AI chooses to cite your content—and that choice is influenced by the grounding queries it formulates.
By examining those queries, you can answer questions that have been opaque until now:
- Is my content considered authoritative enough to be cited? Low citation counts for a high-volume topic may indicate the model doesn’t trust your source, or your content doesn’t align with the grounding queries the model prefers.
- Am I targeting the right search intent? The grounding queries often reveal the model’s deeper, decompositional intent. You might find that while your article targets “best laptops,” the model’s queries are more specific, such as “laptops with long battery life for students 2024.” This mismatch means you’re missing citation opportunities.
- Are there entirely new user needs I should cover? Grounding queries expose search angles you might never have considered. For example, a recipe site might discover that Copilot is querying “gluten-free chocolate cake without xanthan gum” far more often than the generic “chocolate cake recipe.” Creating content specifically for those long-tail, grounded intents can increase the chances of citation.
Essentially, the AI Citations dashboard gives SEO practitioners a new lens: not what people are searching for, but what the AI is searching for when trying to answer people.
How grounding queries differ from traditional keyword data
| Dimension | Traditional Search Console Keywords | Grounding Queries in Clarity |
|---|---|---|
| Source | User-typed queries on search engine | AI-orchestrated search strings |
| Intent representation | What the user explicitly asked | What the AI interprets and decomposes the task into |
| Granularity | Single queries, often broad | Often multiple sub-queries per user prompt |
| Contextual enrichment | None; isolated search | Grounding queries are part of a conversation chain |
| Reporting depth | Impressions, clicks, position | Citations, clicks, page-level drill-downs |
Because the grounding queries are machine-generated, they may include unnatural phrasing, concatenated terms, or filters that a human would never type. You might see a query like “fast laptop under 800$ weight<3lbs 2025” which, while odd to a human, perfectly triggers a set of caching layers and ranking algorithms. Ignoring these patterns means leaving citation share on the table.
Interpreting the data: a practical framework
When you first open the AI Citations report, the sheer volume of queries can be overwhelming—especially for large sites. Here’s a method to extract actionable insights:
-
Filter for high-citation queries: sort the Grounding Queries table by citation count descending. These are your superstar topics where the AI already finds your content useful. Study what these pages have in common: structure, depth, multimedia, authoritative links. If a competitor is cited more often for the same queries, audit their content to understand why.
-
Identify content gaps: look for grounding queries that cite multiple pages from other domains but never yours. These queries likely correspond to a page you don’t have, or one that is too thin. Flag these as content opportunities and prioritize them by citation volume.
-
Monitor declining citations: if a page’s citation count suddenly drops, the model might have switched to a newer source or deem your content less relevant. This is analogous to a ranking drop—act quickly to refresh, add new data, or improve E-A-T signals.
-
Cross-reference with Clarity’s user recordings: open a few session recordings filtered by the AI citation referral. How do users interact with your page after clicking a Copilot citation? Are they scrolling deep, converting, or bouncing back to the AI? This behavioral data helps you optimize the page for post-click engagement, which may indirectly influence future citation likelihood.
Limitations and what Clarity doesn’t show
The AI Citations dashboard is not a panacea. Several blind spots remain:
- No impression data: you see only successful citations, not how many times your page was considered but rejected. You’re still missing the “position zero” equivalent.
- Query transformation logic is hidden: you see the final grounding query, but not the intermediate chain-of-thought steps that produced it. Understanding why the model chose that particular string often requires manual reverse-engineering.
- Geographic and personalization filters: citations can be influenced by a user’s location, language, or prompt history, but Clarity aggregates all citations without segmenting by these dimensions.
- No direct path to the original user prompt: a grounding query like “top running shoes for flat feet men” could have originated from dozens of different conversational contexts. Tying it back to the full user intent is still largely inferential.
Despite these gaps, the data is a massive leap forward from complete darkness.
The broader shift: from ranking to grounding
For more than two decades, SEO was about convincing a static algorithm that your page deserved the top spot for a given keyword. The paradigm is shifting to a dynamic, dialogic system where the algorithm asks its own questions. Visibility is no longer a rank; it’s a retrieval and citation probability.
This has profound implications for how content creation teams operate:
- Topic depth over keyword density: because grounding queries can break a complex question into many facets, pages that comprehensively cover a subject are more likely to be retrieved for multiple sub-queries. Thin, keyword-stuffed pages will rarely satisfy an expansive grounding chain.
- Structured data is non-negotiable: schema markup, especially FAQ, HowTo, and Article schemas, help models parse and cite precise snippets. The more machine-readable your content, the more likely it is to be sliced and cited correctly.
- Freshness signals matter: citation data in Clarity shows that the model often prefers recently updated or published content, particularly for time-sensitive grounding queries. Regular content refreshes are becoming a citation driver.
- Authoritative backlinks still count, but differently: the AI may weigh anchor text context and source reputation when deciding whether to cite a page. While you can’t see this directly, backlink profile likely remains a trust signal.
How to enable and access AI Citations in Clarity
If you haven’t set up Clarity yet, it’s free and requires adding a small JavaScript snippet to your site’s <head>. Once tracking is active and data collects, the AI Citations report will appear once Copilot begins citing any page from your domain. No additional configuration is needed. Bing’s indexing team has confirmed that citations rely on the standard Bing index—so ensuring your pages are indexed, crawlable, and have clear metadata is a prerequisite.
A quick setup checklist:
- Install the Clarity tracking code on all pages.
- Verify your site is indexed by Bing (use Bing Webmaster Tools).
- Submit a sitemap and fix any crawl errors.
- Add structured data where appropriate.
- Monitor Clarity’s “AI” menu for the AI Citations report to activate (may take several weeks depending on traffic).
A new era for search transparency
By exposing grounding queries, Microsoft is giving publishers a tool that Google has yet to replicate in any of its AI-powered search experiences. Google’s Search Console still reports only classic search traffic, even as AI Overviews roll out to billions of queries. The Clarity AI Citations dashboard sets a new benchmark for transparency and pragmatism in AI search.
Forward-looking SEO teams should treat this as a research laboratory for understanding how machines read the web. The insights you glean from grounding queries today will inform the content architectures of tomorrow—where being cited isn’t just a vanity metric, but the primary signal of organic success in an AI-mediated information ecosystem.
As Bing Copilot and its enterprise cousins expand, the ability to reverse-engineer machine intent through grounding queries will become a core competency for any digital marketer. The black box hasn’t turned crystal clear, but at least now we have a flashlight.