How this is calculated

Each category is measured on a fixed set of buyer questions — 10 of them — put to 6 AI assistants in 3 markets, every day. On any given day a category receives 4 to 10 of its questions across 4 to 6 of the assistants; the set is fixed, the daily run samples it.

The questions never change. The same set has run every day since the panel’s current configuration began. When a product moves up or down a leaderboard, the answers moved. The questions did not.

They are not asked in equal volume: the bar beside each is its share of the category’s scored answers. Some read like a template, because the category name is dropped into one. Both are properties of the panel, so both are shown rather than tidied.

We put the questions to the AI assistants every day and record what each one said. Of those answers, we score the ones that cited G2 and named at least one product. That count is printed on every leaderboard and is the denominator of every metric. The real boundary is the panel itself: a fixed set of questions per category, not every question a buyer could ask and not the whole internet.

Each leaderboard also reports a sharper figure: how many scored answers cited that category’s own page rather than a product page, a comparison or a discussion thread. G2 serves every category at two addresses — /categories/live-chat and /best-live-chat-software — and AI cites both, in proportions that differ sharply by category, so an answer counts there if it drew on either.

A vendor comparing this to their own Profound or Semrush dashboard should expect a different figure, because it answers a different question: not how often does AI mention you anywhere, but when an AI answer is built on G2’s view of your category, where do you place.

Questions run daily; the leaderboard is rebuilt weekly over a trailing 30 days — currently 5 October 2026 to 3 November 2026, pulled 4 November 2026. Nothing inside the window is weighted by recency: an answer from the first day counts exactly as much as one from the last.

So a rank only moves when the answers move. New answers enter the window and the oldest drop out, and coverage, prominence and top-3 rate shift with them. That is the movement the score exists to show, and it is the only thing that produces it.

AI answers name brands, not G2 listings, so every name is matched to a product on the category’s Grid before it counts. A category is published only where most of what AI names can be matched, because a ranking that silently ignores a large share of the names is worse than no ranking.

Two of the matching rules are judgment calls, so their share is shown for every category: a company’s name credited to its only product on the Grid, and a name two Grid products share credited to the one with at least three quarters of the reviews between them. Where neither holds, the name is left out.