How to read Microsoft Clarity's new branded vs. non-branded AI citations

Most AI-visibility reporting has the same structural flaw, and it is not a small one: it reports a single mention-rate or citation-share number and lets you assume it means you are winning the category. It might mean that. It might also mean an AI system is simply repeating your brand name back to people who already typed it into the prompt. Those are two completely different outcomes, one is durable competitive advantage and the other is an echo, and until three weeks ago there was no first-party tool that separated them cleanly enough for a marketing team to act on the difference with any confidence.

What actually shipped on August 3

Microsoft Clarity's engineering blog, under a Clarity Staff byline, published the update on August 3, 2026: branded query segmentation added directly to the existing AI Citations dashboard. Every grounding query Clarity tracks, the queries AI systems run against the open web when composing an answer that references your site, now gets tagged as branded or non-branded. A branded query is a direct mention: someone asked an AI system about your company or product by name, and it went looking for grounding. A non-branded query is category-level: someone asked a general question in your space, with no brand named at all, and an AI system chose, independently, to retrieve and cite you as part of the answer.

Clarity frames this as the natural next step after two prior citation-side releases in the same 25-day window, which is itself worth noting: Microsoft is iterating on AI-visibility measurement faster than most GEO vendors are, and it is doing it inside a free analytics tool that already sits on a large share of enterprise sites. That distribution matters more than any single feature. A capability that ships free, inside a tool teams already have installed, reaches production dashboards faster than anything requiring a new vendor evaluation.

Why the blended number was hiding the real question

We have written before about how a single blended mention-rate dashboard can actively mislead you, because it hides engine exclusivity, ghost mentions, and retrieval undercounting behind one reassuring topline figure. The branded/non-branded split attacks a different, related blind spot: even a mention-rate number that is technically accurate can still be answering the wrong question, because it doesn't tell you what kind of citation you're counting.

A high branded-citation share with a thin non-branded share describes a company with strong existing brand recognition and weak category-level discoverability: people who already know you get confirmed, people who don't know you yet never encounter you through an AI answer. A high non-branded share is the harder, more valuable signal. It means an AI system is choosing you, unprompted, when someone asks a general question in your space with no brand name attached at all, which is the AI-search equivalent of ranking for a competitive, non-branded keyword instead of just showing up for your own company name.

Walking through a real reporting scenario

Here is how this plays out on an actual reporting call, because the abstract version undersells how different the two numbers can look for the same site. Say a mid-market B2B SaaS company pulls its Clarity AI Citations dashboard for the first time since the split shipped, expecting the blended citation count from the July launch to hold up. It doesn't split evenly. The dashboard shows the large majority of citations are branded: AI systems reliably surface the company's own pages when someone asks about the company by name, which is a low bar to clear and mostly reflects that the brand already exists and has an indexed, functioning site.

The non-branded share, filtered separately, tells a worse story: thin, concentrated in one or two topic clusters, and essentially absent everywhere else the company competes. That is the moment a branded-only mention-rate report would have missed entirely, because the blended number looked fine. The correct read is not "we're doing well in AI search." It's "we're doing well at getting recognized by people who already know us, and invisible to everyone else's first question," which is a genuine strategic gap, and one a single citation-share number actively conceals.

Now run the comparison against a direct competitor in the same category, someone with a smaller brand footprint but a more deliberate GEO program. Their branded share is lower in absolute terms, unsurprising given the awareness gap, but their non-branded share is close to double. Pull that pair of ratios into the same slide and the story changes completely from what either company's raw citation count would suggest on its own. The bigger, better-known brand is coasting on recognition. The smaller one is actually winning the category conversation on merit, one unprompted citation at a time, and it is the number that predicts where the gap closes next, not the one that describes where things stand today.

It is worth being clear about what kind of measurement tool Clarity is in this picture, because it is easy to mistake one useful dataset for a complete one. We mapped the full measurement picture in the four datasets behind real AI visibility measurement: server logs, first-party citation dashboards like Clarity's, third-party AI-visibility platforms, and branded-search lift modeling. Clarity's branded/non-branded split is a genuine improvement to the second dataset in that list. It is not a substitute for the other three, and a team that reports Clarity's number as the whole picture has just traded one incomplete metric for a slightly less incomplete one, which is progress, but not the finish line some vendors will imply it is.

Where this data still runs out

This feature is genuinely useful and it is also not a complete measurement system, and Clarity's own documentation does not claim otherwise. Grounding-query detection depends on Clarity's visibility into the specific AI systems it tracks, which is not every engine your buyers use, and the branded/non-branded classification is a heuristic applied to query phrasing, not a verified read of user intent. A borderline query, one that mentions a competitor's product category using your company's terminology, can land on either side of that split in ways worth spot-checking rather than trusting blindly. Treat the first month of data as a baseline to sanity-check by hand, not a number you forward to leadership unread.

It also still doesn't solve the click and conversion attribution gap that runs through every first-party AI-visibility tool shipped so far this year. Clarity's citation dashboards, like Search Console's generative AI report, count frequency: how often you show up, and now, how often that showing-up was earned versus recognized. None of them yet report what a searcher did after that citation appeared, whether they clicked, whether they converted, or whether the citation influenced a purchase decision that showed up as a branded search three weeks later with no referrer at all.

Building it into a monthly reporting cadence

The practical move is to stop reporting a single AI-citation number and start reporting two, every month, next to each other. Branded share tells you whether your existing brand demand is being reflected back correctly. Non-branded share tells you whether you are winning new category territory you didn't already own. A GEO program that only tracks the first number can look successful indefinitely while doing nothing to expand who discovers the brand in the first place, which is the entire point of showing up in an AI answer instead of just a branded search result.

This pairs directly with the work we do inside reporting and analytics engagements, where the recurring problem is not a lack of data, it's a dashboard that reports one number when the underlying reality requires two, sometimes three, before anyone can act on it responsibly. Clarity's split is the first time that particular pairing has shipped for free, inside a tool most enterprise marketing teams already have running, which removes the last excuse for reporting AI visibility as a single, misleadingly reassuring line on a slide nobody questions.

Do this next: open your Clarity AI Citations dashboard, filter by branded and non-branded separately, and compute the ratio for your last 30 days. If you don't like what the non-branded number says, that's not a measurement problem to fix. It's the actual list of topics your content and technical GEO work need to target next, handed to you for free by a filter that didn't exist three weeks ago. Run the same split for your two or three closest competitors while you're in there, using whatever domain-level visibility your existing tools already expose, and you'll usually find the gap is not in total citation volume at all. It's in which half of that volume is doing the actual work of winning new attention, versus the half that was always going to show up regardless of what you built.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.