Microsoft Clarity now splits AI citations by brand

Microsoft has added branded and non-branded query segmentation to Clarity’s AI Citations dashboard, letting brands see whether an AI engine cited them off a query that named the brand or off a broader, category-level one. These are the grounding queries the AI itself runs to gather sources, not the words a person typed. Microsoft shipped the update on 3 August 2026, its third change to Clarity’s AI analytics in under a month, and it is free. For teams trying to measure how a large brand surfaces in AI answers, it turns a single visibility number into two signals that mean very different things.

What happened

Clarity Citations shows the grounding queries an AI system runs before it cites a site, the searches behind the answer rather than the answer itself. According to the Microsoft Clarity blog, the 3 August update labels each of those queries as branded or non-branded, splits the Share of Authority card by query type, and applies the same filter across the dashboard. Branded queries name the brand directly; non-branded queries are the generic, category-level questions an AI runs to find supporting information.

Microsoft framed the problem the split solves. “Without a clear way to separate those signals, it can be difficult to understand whether your performance reflects direct brand visibility, category discovery, or a mix of both,” the company wrote.

Clarity Citations reached general availability on 13 May 2026, and this is the third addition since, following Topic Insights on 9 July and query-topic analysis on 22 July.

Why it matters

“By separating brand-led demand from generic discovery, you can better assess brand strength, identify discovery and consideration opportunities, and interpret changes in citation performance with greater confidence.”

Microsoft Clarity Blog

AI answers are becoming a real discovery surface, and until now most tools reported a single “you were cited” figure. That number hides the question that matters: did the AI reach for you on a grounding query that named the brand, or on a broader, unbranded category query? The two call for different work. Strong branded citations show you perform when the AI is already looking you up by name; strong non-branded citations show you are surfacing in open, category-level answers where no brand was named. Being able to filter the two apart, in a free tool, lowers the bar for measuring AI search visibility at all.

What this means for multi-location brands

For a brand running hundreds or thousands of locations, the branded and non-branded split maps onto two different jobs. Branded grounding queries are largely a reputation and accuracy question: when an AI looks you up by name, is it pulling correct, consistent information about every location, or stale hours and wrong addresses? That is a reputation and profile accuracy problem. Non-branded grounding queries are the generative engine optimisation question: when the AI fields a prompt like “a pharmacy open now near me” or “EV charging on the A6”, does your estate show up at all?

Clarity reports for a website domain, not per physical location, so it tells a central team where the brand stands overall, not which branch’s data is wrong. That gap is the familiar one: measurement points to the problem, fixing it happens across the estate. If non-branded citations are thin, the answer is rarely one page, it is the underlying data, and clean, structured location data across every profile is what AI engines read before they cite anyone. If branded citations pull the wrong details, the fix is consistent listings management at scale rather than a per-location scramble.

Clarity is worth adding to the measurement stack precisely because it is free and first-party. Treat it as the read on whether the work is landing, and keep the work itself, the data and the listings, in a system built for the whole network.

The bottom line

Splitting AI citations into branded and non-branded turns a vanity metric into a diagnostic. For multi-location brands, it separates “we are known” from “we are found”, and the second is where the harder, estate-wide work lives.