Review Methodology

Every AI companion platform we review is scored across 9 weighted categories on a 10-point scale. The overall score is the weighted average. Platforms we have not hands-on tested do not receive a score.

Version

1.2

Last updated

2026-09-10

First published

2026-06-01

The nine categories

Weights

Total weight adds up to 100. Weights are the single source of truth — any change here automatically propagates through every review and best-of page.

Category	Weight
Conversation quality	15%
Memory	15%
Character customization	15%
Image quality	15%
Video quality	10%
Voice / calls	10%
User experience	5%
Value	10%
Privacy / transparency	5%

How we score each category

Conversation quality — depth, coherence, tone control, roleplay handling, refusal behaviour.
Memory — recall of prior conversations across sessions, days and weeks; how gracefully old context is surfaced.
Character customization — how deeply personality, appearance, voice and roleplay parameters can be tuned.
Image quality — resolution, aesthetic quality, and same-character consistency across generations.
Video quality — motion coherence, character consistency across frames, prompt adherence.
Voice / calls — voice quality, latency, turn-taking, and (if supported) call quality.
User experience — onboarding, navigation, gallery UX, mobile behaviour, edge-case bugs.
Value — pricing vs feature depth vs cap on free tier. Not the cheapest — the best price-for-feature.
Privacy / transparency — clarity of data handling, moderation policy, billing behaviour and account controls.

What counts as "tested"

A platform is "tested" only when we have a paid or free-tier account and have run our standard test protocol against it: multi-session conversation, image generation batches, voice reply samples, feature-matrix verification, a pricing audit, and a cancellation test. Any platform without this is marked "Not yet independently tested" and receives no score.

Retesting

Reviews are retested at least every 90 days and immediately after any major platform update. The date on each review reflects the most recent retest.

Methodology change log

This methodology has changed since launch. Every version shipped is documented below so readers can see exactly what changed, when, and why a score written under one version might read differently than a score written under the next.

Version 1.2 · 2026-09-10

Added mandatory hands-on testing narrative on every published review — testers now write in first person about what the account experience was actually like.
Every review is now credited to a named Person author with a public profile.
Sources / citations are now required for factual claims about pricing, privacy, and features.

Version 1.1 · 2026-07-15

Introduced a mandatory cancellation-and-resubscription test as part of the Value score.
Split "AI features" into the current Conversation, Memory, and Character customization categories for finer resolution.

Version 1.0 · 2026-06-01

Initial nine-category weighted methodology published.
Established the rule that no platform can carry a score without a paid or free-tier account being used.