Methodology
How we review.
Every app gets the same 5-axis scorecard, tested hands-on for a minimum of thirty messages, and re-scored when a meaningful update ships. This page documents the rubric, the protocol, and the editorial independence policy in full.
The rubric
Five axes. One scorecard.
Each axis is scored 1–10. The overall score is a weighted average rounded to one decimal place. Weights reflect what changes a real session most.
| Axis | Weight | What it measures |
|---|---|---|
| Chat quality | 25% | How well the AI holds conversational rhythm, mirrors the user’s tone, maintains character voice across topic shifts, and avoids template-like output. Measured across at least 30 messages per platform, across casual chat, roleplay, and emotional context. |
| Customization | 20% | Depth and usefulness of persona controls (appearance, personality, chat style, scenarios, response length, memory). We weight controls that meaningfully change session feel over cosmetic toggles. |
| Visuals | 15% | Quality and consistency of static character art and, where supported, dynamic image generation. We assess prompt fidelity, identity coherence between chat and image, and whether visuals enhance or distract from the chat experience. |
| Pricing value | 20% | Real cost relative to delivered quality, free-tier usefulness, and whether premium gates land on essential or optional features. Pricing is benchmarked against the median of the curated set, not the cheapest competitor. |
| NSFW freedom | 20% | How permissive the platform is with mature roleplay, escalation pacing, and explicit content - without compromising user-safety controls. Platforms designed as SFW are not penalised, but reviewed within their own category. |
The protocol
Tested by humans.
- Minimum thirty messages per platform, across casual chat, roleplay, and emotional context.
- Sessions split across at least two days to test return-visit memory.
- Identical persona prompts used across platforms where supported, so persona depth is comparable.
- Pricing benchmarked against the median of the curated set, not the cheapest competitor.
- Reviews refreshed when significant features ship, with the last-updated date stamped on every page.
Editorial independence
Affiliate links never set the order.
CompareAIGF is independently owned. We accept no payment for placement, ordering, or score adjustments. Affiliate relationships exist for some of the platforms we cover - those links are disclosed on every review page - but do not influence scoring, ranking, or whether a product makes the curated set.
Where a partner platform sits underneath an affiliate funnel (for example, a lander funnelling to a destination app), we make that relationship explicit in the review body so readers can weight the recommendation appropriately.
See our full disclaimer for the affiliate programmes we participate in.
Corrections
Found something off?
Methodology feedback and correction requests are welcome via the contact page. Reviewers are happy to walk through scoring decisions on request.