bestaiq
Home / Methodology
// METHODOLOGY

How we score AI tools

One transparent rubric, applied the same way to every tool. We buy the subscriptions, run identical tests, and document every number — so a score means the same thing across the site.

FIG · IN ONE PARAGRAPH

Each tool earns a 0–10 score from six weighted dimensions — capability, value, ease, privacy, support and ecosystem — then a documented editorial adjustment for risks the formula under-weights. Weights are tuned per category (image, video, code). We hands-on test every tool on an identical task set. There is no paid placement, and affiliate links never move a score.

FIG · HOW A SCORE IS BUILT
Every verdict shows its work
SOURCES
47
SUB-SCORES
6 DIMS
WEIGHTED
Σ=1.0
EDITORIAL
+OVERRIDE
VERDICT
6.9/10
SOURCES
Official pricing, docs, regulatory filings and independent benchmarks — every claim tagged.
SUB-SCORES
Hands-on scoring of all six dimensions from the same task set.
WEIGHTED
Sub-scores multiplied by category-tuned weights that sum to 1.00.
EDITORIAL
A documented adjustment (±0.5 cap) for risks the formula under-weights.
VERDICT
A single 0–10 score plus its tier band — the same meaning site-wide.

//The six dimensions

0.30Capability

Output quality, fidelity and reliability/consistency.

Rationale — the core of what you pay an AI tool for → highest weight.
0.20Value

Price relative to real output (credits, caps, tiers).

Rationale — AI pricing varies wildly, so value is decisive.
0.15Ease

Onboarding, UX, time-to-first-result.

Rationale — friction is a real adoption barrier.
0.15Privacy

Data handling, security, training opt-out and safety/compliance.

Rationale — rising stakes (enterprise, EU AI Act); weighted higher for code.
0.10Support

Docs, support, community, changelog.

Rationale — matters but rarely decisive.
0.10Ecosystem

Integrations, API, platform reach.

Rationale — a multiplier on the others, not a core.
TABLE · DIMENSIONS & BASE WEIGHTS
·Σ= 1.00

Priorities differ by niche, so weights are tuned per category (each column sums to 1.00). Example: privacy weighs more for code tools; output quality weighs more for image.

DimensionBaseImageVideoCode
Capability 0.30 0.35 0.35 0.28
Value 0.20 0.20 0.22 0.15
Ease 0.15 0.15 0.13 0.12
Privacy 0.15 0.10 0.08 0.20
Support 0.10 0.05 0.07 0.10
Ecosystem 0.10 0.15 0.15 0.15
Σ Total1.001.001.001.00
TABLE · WEIGHTS BY CATEGORY

Profiles proposed v1.0 — reviewed quarterly.

//The formula

SCORE ENGINE v1.0 · ±0.5 CAP
SCORE = Σ(dimension × category-weight) + editorial override
WORKED EXAMPLE · LEONARDO.AI (IMAGE)
Leonardo.Ai = 8.0×0.35 + 6.0×0.20 + 7.0×0.15 + 5.0×0.10 + 9.0×0.05 + 8.0×0.15 = 7.10 −0.20 editorial (privacy risk) 6.9/10

The editorial override is always documented per tool and is capped at ±0.5.

//Score bands

9.0–10 Editor’s Pick T1
8.0–8.9 Excellent T1
7.0–7.9 Great T2
6.5–6.9 Good T2
5.5–6.4 Fair T3
4.5–5.4 Mixed T3
3.5–4.4 Weak T4
0–3.4 Avoid T4
TABLE · SCORE BANDS

//How we test

We create free and paid accounts with our own money, run the same tasks on every tool, screenshot the results, and record the date. We never accept vendor-supplied demos or sponsored access.

TESTED WITH: Free tier· Paid tier· Jun 2026
FIG · THE TEST BATTERY

Rubrics are category-specific: each tool type runs the task set built for its niche, scored against the same rubric across every tool in that category.

IMAGE

5 identical prompts run on every generator

  • 01Photoreal portrait, 85mm, controlled soft window light.
  • 02Vintage travel poster with a legible, correctly-spelled title.
  • 03Isometric product scene — consistent style across a 4-image series.
CODE

Identical tasks + SWE-bench-style benchmark

  • 01Fix a failing test in a real open-source repo (SWE-bench-style).
  • 02Refactor a 300-line module and keep the public API unchanged.
TABLE · SOURCE PRIORITY
01 Official / pricing pages
02 Product documentation
03 Regulatory / legal
04 Independent benchmarks
05 Reputable review sites (G2, Trustpilot)
06 Community (Reddit)
07 Marketing / press

Lower-priority sources never override higher ones.

//Editorial independence & affiliate policy

Independent · ad-free verdicts · we may earn affiliate commissions — this never affects our scores.
B22 · CHANGELOG
Version Date What changed
v1.0 Jun 2026 Initial public methodology — six-dimension rubric, category weight profiles, ±0.5 editorial cap and score bands published.
v0.9 Apr 2026 Internal calibration pass: added privacy safety/compliance sub-axis and re-balanced code-category weights.

Re-verified quarterly · full audit annually.

//Frequently asked

Q1

Do affiliate links affect your scores?

No. Scores and rankings are computed from the rubric before any commercial relationship is considered. Affiliate links use rel="sponsored nofollow" and never move a score or a rank — this is a hard editorial rule.

Q2

Why do weights change by category?

A code assistant and an image generator are not judged on the same priorities. Privacy and compliance matter more for code that touches your repos; raw output quality matters more for image. Each category profile still sums to 1.00, so scores stay comparable within a niche.

Q3

Why a 0–10 score and not 5 stars?

A 0–10 scale with documented bands gives us the resolution to separate a 6.9 from a 7.4 — differences a 5-star scale collapses. Every score also carries its tier word (Good · T2) so the number is never read alone.

Q4

Do you accept paid placements or sponsored reviews?

No. We do not accept payment for placement, ranking, or a favourable review, and we never publish vendor-supplied or sponsored content as editorial. We only partner (via affiliate links) with tools that already clear a minimum score on their own merits.