How we test and score AI tools
Principles
- Independence. Anchors, weights and gates are fixed before scoring and identical for every tool in a category. Rankings are never sold and never weighted by affiliate payout.
- Evidence over vibes. Every dimension score is tagged with how it was obtained. If we did not measure something, the card says “not measured” — it never gets a guessed number.
- An honest ceiling. No tool gets a perfect score, by design. Top picks land in the mid-to-high range.
- No covering up weak spots. We combine dimensions with a rule that will not let a brilliant capability score erase a documented trust problem. A well-evidenced consumer-trust failure caps the composite outright.
- Freshness. Every score is pinned to a collection date and the tool version measured. Aged scores are marked stale and re-collected.
The five dimensions
| Dimension | What it answers, and where the number comes from |
|---|---|
| Capability | How good is the core output? Sourced from public output-quality arenas — thousands of blind pairwise votes — or, when we run one, our own controlled test battery. |
| Reliability | Do you get a good result consistently, not just on a lucky try? Only a controlled repeat-run can measure this, so it is marked “not measured” until we run one — we never guess it. |
| Usability & control | How easy is it to get what you want? Aggregated from capability-focused review crowds (G2, Product Hunt, Capterra), weighted by sample size. |
| Value | What does a usable result actually cost? Verified official pricing, free-tier availability, billing terms. |
| Safety & trust | Commercial/IP safety and whether the vendor behaves: copyright indemnification terms plus consumer-trust records (billing, refunds, support) from Trustpilot. |
Bands are published so a number means the same thing everywhere: 90–100 exceptional · 75–89 strong · 60–74 adequate · 40–59 weak · below 40 poor.
Evidence grades
Every dimension carries a grade for how we know:
- A — controlled test. We ran a repeatable benchmark ourselves.
- B — measured public data. Blind-vote arenas and review aggregates (G2, Trustpilot, Product Hunt, Capterra) with sample sizes and dates disclosed. Sources are routed to the dimension they actually measure — billing complaints never contaminate a capability score.
- C — vendor documentation. Official pricing and legal terms, verified on a stated date.
Thin or contested evidence lowers the published confidence; a vendor’s unverified claim can never lift a score on its own.
From dimensions to one number
Dimension scores are normalized against absolute anchors — so a rival’s launch never moves a tool’s score — and combined with a weighted geometric mean, which punishes lopsided profiles harder than a simple average would. Critical-dimension gates then bite in proportion to evidence: a trust score below 50 backed by 100+ reviews hard-caps the composite; the same signal from a thin sample only flags it. Finally the honest ceiling caps everything below a perfect score. The composite is secondary output — the five-dimension profile with its confidence is the review.
Version and disclosure
This page describes Trust Score v1. The methodology is versioned; every published score states the version that produced it. We are reader-supported and may earn an affiliate commission when you buy through our links, at no extra cost to you — rankings are never sold, and paid submissions buy review speed, never a listing or a score.