Microsoft Clarity AI Citations: What It Tracks (and What It Misses)

Microsoft Clarity now reports which of your pages AI answers cite, for free, using data you cannot get from prompt tracking. The Citation dashboard went generally available on 13 May 2026, and it is built from the citation logs of the systems that generated the answers rather than from sampling. That makes it genuinely new. It also sees one ecosystem, which is the part most coverage skips.

Key takeaways

Clarity's Citation dashboard sits under AI Visibility and reports five things. Microsoft's definitions are precise enough to be worth taking literally.

Two notes in the documentation do more work than the metric definitions. The first sets the scope: "The Citation dashboard tracks citation activity in AI-generated answers only. It does not measure traditional search rankings, impressions, or click-through rates." The second is a warning against reading citations as placement: counts "reflect how often a page was referenced, not its ranking or prominence within an AI-generated answer." Being cited third in a hedged paragraph and being the source an answer is built on count the same.

Setup is a one-time domain verification by a project administrator, through the Clarity tracking code, Google Search Console, or Bing Webmaster Tools. On 9 July 2026 Microsoft added Topic Insights on top, grouping citations into topics with content recommendations. It ships in beta, and Microsoft's own framing is that it "should only be used for directional monitoring and insights."

Here is the distinction that makes this worth your attention. There are two ways to know whether AI answers cite you, and they answer different questions.

Observed data comes from the logs of the system that generated the answer. Clarity can tell you your pages were cited 400 times last week because Microsoft's infrastructure did the citing. There is no sample, no estimate, and no prompt set to argue about.

Sampled data comes from running prompts yourself and reading what comes back, which is the method behind prompt tracking and share of voice. It is an estimate, and it is engine-agnostic: you can point it at anything with an API or an interface.

Neither supersedes the other. Observed data settles volume questions that sampling can only approximate. Sampling answers the questions a log cannot: what the engine said about you, whether it got your pricing right, and which three competitors it recommended instead. Our study of citation volatility is an example of the second kind, and it also explains why a single sampled check is weak evidence: across 28 prompts tracked daily for six weeks, the set of cited domains turned over by roughly 60–70% day to day.

Microsoft describes Citations as surfacing activity "across supported AI experiences" and does not publish the list. What the product does disclose is where the data comes from: it unifies signals from Bing Webmaster Tools and Clarity, spanning crawling and indexing through to citation and AI-referred traffic. That is Microsoft's own retrieval stack.

The practical read, as Search Engine Journal put it in May 2026, is that Clarity captures citations "primarily across Microsoft's AI surfaces (like Copilot and Bing generative search)" and "is not giving you a direct window into how OpenAI's ChatGPT (using its own search), Google Gemini, or Perplexity are citing your links, because those platforms do not share their internal grounding logs with Microsoft."

That is not a flaw in the tool; it is the structural condition of first-party AI data. Every platform can only report on its own retrieval, and none of them will report on each other's. So a dashboard fed by one ecosystem's logs is precise and narrow at the same time. Treat the number as accurate for what it covers, and do not let it stand in for AI citations generally.

Share of authority looks like a share-of-voice metric and behaves slightly differently, in a way Microsoft documents openly. The calculation runs at a daily level: "If your domain is cited for a query on a given day, all citation instances for that query on that day are included; if not, that day is excluded." Microsoft then says what follows from that, which is the useful part: the approach "focuses on active participation and may lead to higher Share of authority values compared to broader calculations."

In other words, days you lost are not fully in the denominator. A 30% share of authority is not the same claim as "we appear in 30% of relevant AI answers." It measures your slice of the citations on the queries and days where you were already in the running, which is a narrower and more flattering base.

That makes it a good trend line and a bad boardroom number. Watch it move, and pair it with a presence-style measure across your actual prompt set if you need to answer "how often do we show up at all?"

The metric with the most to teach you is the least headline-friendly. Grounding queries are the phrasings the engine used to retrieve, and Microsoft is upfront that they "may differ from the exact phrases users type." A model takes a conversational question, decomposes it, and issues its own retrieval terms, which is the mechanism that makes grounding work at all.

Seeing those terms is close to reading the engine's notes. You learn which of your pages satisfied which retrieval intent, where a single page is absorbing queries that deserve their own coverage, and which adjacent phrasings retrieve competitors instead of you. That is a concrete content brief, and it is the kind of input answer engine optimization usually has to infer.

The caveat is the same one as above. These are one ecosystem's retrieval patterns. They generalize reasonably, because engines that ground answers face similar retrieval problems, but "generalizes reasonably" is an assumption, not a measurement.

If you are assembling AI visibility measurement without a budget, Clarity is a real addition rather than another grader. It gives you observed citations, grounding queries, and AI referral traffic for nothing, which no free prompt-based option can match on precision.

It leaves three gaps: the engines outside Microsoft's stack, the wording and sentiment of the answers themselves, and the competitor set named beside you. Those need sampling. The method is in our guide to tracking your brand in AI search, and Elmo automates it as an open-source, self-hosted platform that runs your prompts across ChatGPT, Claude, Gemini, Perplexity, and Google's AI surfaces and records every mention, citation, and rival named alongside you. For the rest of the no-cost options, see our roundup of free AI visibility tools; for the paid category, the best AI visibility tools comparison and the AI visibility software hub.

The short version: turn Clarity on, because free observed data about your own citations is worth having. Just do not mistake one ecosystem's logs for a view of AI search.