Search moved from the ten blue links to one paragraph. cited asks your market's buying questions to real AI engines, counts every entity that gets named, and hands you the receipt: who was named, where you rank, what to fix, and re-measurement after you fix it.
$ cited scan --brand you --market "your market" --competitors "them"sweap.it, our own memory-hygiene tool, put against the npm dependency management market: 16 buying questions, one real AI engine. The receipt came back ugly. That is the point: a measurement instrument that flatters its owner measures nothing.
| Entity | Mentions | Share of voice |
|---|---|---|
| sweap | 0 | 0% |
| renovate | 19 | 59% |
| dependabot | 13 | 41% |
"For npm projects, the core tools are npm, Yarn, and pnpm. […] For updating dependencies, use npm-check-updates (CLI) or Renovate (automated PRs). Dependabot (GitHub native) is also solid for security updates. […] Start with pnpm + Renovate for a modern setup." answer 1 of 16 · counts: sweap 0 · renovate 2 · dependabot 1
32 competitor mentions, zero ours. The fix list is already written, and the next scan either shows the number move or it does not. Every claim on this page is reproducible: same questions, same market, one command.
Three moves, one artifact per scan: the receipt.
20 buying questions per scan across 7 intent categories: best-of, alternative, vs, use-case, budget, migration, problem. Wording is pinned and neutral. Your brand never appears inside a prompt.
Every answer parsed with boundary-safe, capitalization-aware matching. Notion the product counts; "the notion that" does not. Common-word names get their own guard, so a frequent English word cannot inflate your score.
Fix the gap, run the scan again, diff the receipts. Visibility is a number that moves, not a vibe. Dry runs preview the exact questions for zero cost before a single call is made.
A mention rate from 16 answers is an estimate, and printing "33%" without error bars is false precision. Every headline number on a cited receipt ships with its sample size and a 95% Wilson interval.
Zero after 16 answers is not proven zero: our own scan's interval reads up to 19%. The receipt says what the data can and cannot claim. Engine failures become rows, never crashes. A hard cap refuses any scan beyond 200 calls.