Research methodology
Our frameworks and category pages rest on a repeatable research process. This page documents how we collect data, what we measure, and where our methods have limits.
Last updated 2026-06-04
Pattern sampling
For each industry we maintain a working corpus of 200–500 active brand names sampled from acquisitions, funded startups, marketplace top performers, and editorial best-of lists. The corpus is refreshed annually.
From the corpus we extract structural attributes — syllable count, vowel patterns, prefix and suffix conventions, real-word vs invented vs compound construction — and surface the patterns that recur above a meaningful threshold.
What we measure
- Brandability: distinctiveness, pronounceability, spelling stability.
- Availability heuristics: domain morphology that correlates with .com availability.
- Category fit: whether a name reads as belonging to its industry.
- Trademark hygiene: structural attributes that correlate with cleaner USPTO classes.
What we explicitly do not measure
We do not score subjective brand emotion or cultural appropriateness with a model — those decisions require human judgement and cultural context. We surface candidates and the reasoning behind them; final calls belong to the operator.
Limits and disclosure
Naming corpora are biased toward English-speaking markets and toward brands that have already succeeded — survivorship bias is real. We say so where it matters and we caveat recommendations for non-English markets.