Why CRM, project management, email marketing, and help desk: how we built the State of AI Visibility study

The State of AI Visibility study puts a specific number in front of readers — an average score, an "effectively invisible" count, a leaderboard sorted by category — and any number like that deserves scrutiny before you act on it. So this is the short, plain account of how the study is actually built: which brands, why those four categories, and the one rule we hold to that keeps every number in it honest.

Four categories, chosen for contrast, not convenience

The study covers 18 brands across four SaaS categories: CRM, project management, email marketing, and help desk. Those four weren't picked at random — they're categories with real, well-known incumbents (household names most readers will recognize on sight), genuine competitive density (multiple credible options a buyer would actually shortlist, not one obvious winner), and buyer questions that are naturally comparison-shaped: "best CRM for a small sales team," "help desk software for a startup." That combination matters, because a category with only one real player, or one nobody's heard of, wouldn't tell you anything generalizable about how AI engines handle recommendation questions.

Real scans only — a brand with no scan simply doesn't appear

Here's the rule the study enforces mechanically, not just as a policy: every row on the leaderboard reflects a real, completed scan, read live from the same store every free scan on this site writes to. A brand that hasn't been scanned yet — or whose most recent run came back as a mock fallback rather than a genuine engine response — is simply omitted from the page, not shown with a placeholder or an estimated number. That's why the page's own copy is explicit about scans still "queued" rather than guessed. We'd rather publish a shorter, real leaderboard than a complete, partly-invented one.

Same pipeline, same questions, same scoring — no thumb on the scale

Every brand in the study is scanned with the identical 25-question buyer-intent probe used on every free scan, run through the identical scoring formula — same weighting, same exclusion of branded questions, same rejection of disclaimer-only mentions and invented ranks. There's no separate "study mode" that scores differently than what a paying customer sees on their own report. If we hand-tuned the study's numbers to look more dramatic than a real customer's report, the study would be marketing theater instead of evidence — and the first customer who noticed the discrepancy would be right to distrust everything else we publish.

Why the leaderboard updates instead of freezing in time

Unlike a typical "state of the industry" PDF that's accurate on publish day and stale a year later, the study page reads the live store on every request. When we re-scan a curated brand, its row updates the next time someone loads the page — no re-publish cycle, no separate "2026 edition" to write. The average score, the invisible count, and the "most invisible name" callout are all recomputed from whatever the latest real scans say, which means the honest caveat about any snapshot statistic applies here too: a single scan is a sample, and the trend across repeated scans is the more trustworthy signal than any one moment.

Re-check it yourself

The point of explaining the mechanics this plainly is that you shouldn't have to take the 42-average headline on faith. Browse the leaderboard, filter by industry, and open any brand's full report — the same report a customer sees — to see the actual answers behind the score. Then run the same real pipeline on your own domain: 25 buyer questions, four engines, no signup, about a minute. You'll get the same kind of number this study is built from, computed the same honest way.