How We Review AI Companion Services

The scorecard: eight criteria, fixed weights

Scores are 0–10 per criterion, applied to every ai girlfriend app in the comparison by the same desk on the same dates. The overall score starts from the weighted sum; the desk then sets the final number, weighing the criteria that matter most for the service in question — which is why a single weak criterion does not mechanically cap a total. The same eight numbers and the same overall appear on the hub table and in every deep review — if you find one that disagrees, that is a bug, and we want to hear about it.

1. Chat quality — 22%

What it measures: coherence of replies, staying in character across a long conversation, prose that is not padded with filler, and graceful handling of scene shifts. What separates an 8 from a 5: an 8 keeps a character consistent and interesting for hundreds of turns based on what the service itself documents and demonstrates publicly; a 5 produces fluent but generic replies that could belong to any character, or degrades quickly.

2. Memory and continuity — 14%

What it measures: what the service says persists — conversation memory, summaries of past sessions, relationship progression. An 8 publishes concrete memory behaviour (what is remembered, for how long, at which tier) and demonstrates it in public previews; a 5 either documents nothing or resets the moment a session ends.

3. Characters and customization — 14%

What it measures: depth of the pre-made roster, the range of archetypes (realistic, anime, fantasy), and how far the creation tools go — appearance, personality, backstory, speaking style. An 8 lets you build a specific person with real texture; a 5 offers presets with a name swap.

4. Image generation — 10%

What it measures: whether images are offered at all, their published quality and speed, and whether the same character stays recognizable across generations. An 8 produces consistent, well-composed images of your character on demand; a 5 gives one-off images that could be anyone, at slow speeds or tight limits.

5. Voice and video — 8%

What it measures: voice messages, live voice calls, and any video or animated-avatar features, as documented publicly. An 8 offers natural voice with per-character timbre working on mobile; a 5 has robotic TTS bolted on, or nothing.

6. Explicit content freedom — 14%

What it measures: how open the adult-content policy actually is — where it applies, which tiers gate it, whether scenes collapse into refusals mid-conversation, and whether the marketing matches the enforcement. An 8 is consistent and clearly documented; a 5 either hides the policy or enforces it erratically.

7. Free tier value — 10%

What it measures: what a no-payment visitor actually gets — message counts, image tries, character access, watermarking. An 8 lets you genuinely feel the product before paying; a 5 is a demo that ends before you have learned anything.

8. Pricing transparency — 8%

What it measures: are the real prices, billing periods, token systems and cancellation path published in plain sight? An 8 shows the full cost before signup; a 5 buries per-message math behind three clicks of fine print.

What we could not verify (September 2026)

Anything requiring an account: paid-tier message quality at scale, billing behaviour after renewal, in-app purchase flows and push-notification reliability. We also do not verify claims about underlying models beyond what services publish — when a service names a model, we report the naming, not our own benchmark of it. These gaps are stated in the reviews they affect rather than papered over.