Free tool
Boards are published one category at a time. Nothing on this page is generated on request — every number comes from a run we did and recorded, so a category we have not run yet shows nothing rather than an empty scoreboard.
Back to the free toolsEvery AI visibility tool on the market, this one included, starts by counting mentions. It is the obvious measurement and it is genuinely useful: if your brand string never appears in the answer, nothing else matters yet.
The problem starts the moment it does appear. A buying answer usually names four or five products and then either advises one of them or declines to. Those two outcomes are worth completely different amounts to you, and a mention count scores them identically. A brand named in 70% of answers and advised in none of them has a dashboard full of green and a pipeline full of nothing.
The gap is not a rounding error either. Roundups, category pages and “best tools” listicles are the page shapes engines reach for most on buying questions, and their entire job is to name everybody. If your visibility comes from those, you are being read without ever being the conclusion.
It is a classifier, not a writer. It never composes an answer, never explains a category and never decides who was named. It reads one stored answer at a time and returns four typed labels: which single brand the answer steers towards, whether it steers at all or only lists names, how the brand in question appears, and what kind of source the answer leans on.
Typed is the operative word. Asked who an answer recommends, it returns one option from a list we defined, with a probability across the rest — not a paragraph that has to be parsed back into a number. Nothing is parsed, so nothing is mis-parsed, and the threshold for “this counts as a recommendation” stays in our code where it can be changed and re-applied without paying to judge every answer again.
That is why the run is cheap enough to publish with its receipt attached. Four judgments per answer, one request per answer, and a total at the bottom of this page that came off the run rather than out of a spreadsheet.
Mentions and citations are still rules. Your brand is named if the string is in the answer; a page is cited if the engine returned it as a source. No model adjudicates either one, because a customer who sees a number move deserves a better explanation than “the judgment changed”.
The judgment sits beside those rules, never on top of them. The named column on this page is computed by the same code that fills the share-of-voice panel inside CueScout. If the two ever disagreed, the bug would be worth more than the feature.
Next
Named is a matching rule: your brand string appears somewhere in the answer. Recommended is a judgment: the answer tells the reader to pick you. They come apart constantly. An answer that says "Mixpanel has the largest integration catalogue; the rest trade depth for price" names three brands and recommends none of them, and every visibility tool on the market — including the rest of ours — would score that as three mentions.
Each stored answer is put to a decision model, which returns a typed answer rather than prose: which single brand the answer steers towards, whether it steers at all or just lists names, how the brand in question appears, and what kind of source the answer leans on. Four judgments per answer, one request per answer. An answer that only enumerates options credits nobody, which is why the "no one" row is usually the biggest one on the right.
No, and deliberately not. Mentions and citations stay string and URL matching — cheap, deterministic, and explainable to a customer who asks why a number moved. The judgment is a second, separate reading laid beside them. The two are never merged, and the named column here is computed by exactly the same rules as the share-of-voice panel inside the product.
The size of the run: how many stored answers were judged, how many individual judgments that came to, how long the whole thing took wall-clock, and what it cost. All four come off the run itself. If the provider does not report a cost for a run, the page says so instead of showing a dollar figure we did not measure.
The band above the leaderboards says which. The first board is built from the demo workspace — a fictional product-analytics brand with a full prompt library — because it can be published without exposing a customer's category. The method is identical either way: the same answers table, the same rules for named, the same judgments for recommended.
Because roundups name everyone. A brand that appears in every "best tools" answer and is never the one advised is being read, indexed and summarised without ever being the conclusion — which feels like visibility in a dashboard and converts like nothing. That gap is the most useful thing on this page, and it is invisible to a tool that only counts mentions.
Not from this page. The run costs real money per answer, so it is not something an anonymous visitor can trigger. What CueScout does for customers is the part underneath: asking your buying questions across the engines every week and keeping every answer, which is what makes a board like this possible in the first place.