Methodology

## What we measure AI TrustMark separates **verification**, **review scores**, **rankings**, and any commercial relationship. - **TrustMark** — verified status against published criteria (certificate `AITM-YYYY-######`). - **AI Benchmark Score** — comparative performance out of 100 that drives category rankings. ## Star ratings Qualifying reviews require at least email verification (V1). Unverified submissions (V0) never count. We use a Bayesian / confidence-weighted average: `Weighted rating = (v/(v+m) × R) + (m/(v+m) × C)` Verification level may increase weight (email, organisation, customer relationship, outcome). Exact weights are in the versioned configuration for this methodology. ## AI Benchmark Score Components (illustrative maxima): - Customer satisfaction — /30 - Verified implementations — /20 - Evidence quality — /20 - Customer outcomes — /15 - Recency — /10 - Provider transparency — /5 Every ranking score is explainable via this breakdown. Rank **position is never stored or manually set**. ## Verification levels - **V0 Submitted** — email not confirmed; excluded from ratings - **V1 Email verified** - **V2 Organisation verified** - **V3 Customer relationship verified** - **V4 Outcome verified** ## Fraud controls Signed one-time email tokens, rate limits, bot protection, disposable-email checks, duplicate and similarity detection, related-party disclosure, and moderation queues. ## Ranking qualification A category ranking only lists providers that meet the minimum qualifying review count for that methodology version. We do not publish empty doorway or permutation pages. ## Corrections Administrators may correct source data. They cannot assign a rank position or override the calculated Benchmark Score through an administration field.