How We Test & Rate
AI Girlfriend Platforms
Every score on this site comes from structured hands-on testing under a paid subscription. This page explains exactly what we test, how we score it, and how the numbers combine into an overall rating. No black boxes.
Overview
We evaluate every platform across five categories that reflect what actually matters to users: the quality of conversation, image and video generation capability, voice interaction, customer support responsiveness, and whether the pricing is fair and transparent.
Each category is scored independently by Jordan Voss using the structured protocols described below. Scores are set on a 1.0–5.0 scale in 0.5-point increments. Category scores are then combined into an overall rating using a weighted formula, with chat quality carrying the most weight because it is the core function of the product.
Testing happens under a real paid subscription. Free-tier performance is noted but does not drive the core scores — we care about what paying users experience.
Testing at a Glance
- Reviewer
- Jordan Voss
- Subscription tier
- Entry-level paid plan
- Sessions per platform
- Minimum 3 sessions
- Support tickets submitted
- 3 per platform
- Image prompts run
- 20 standardised prompts
- Video clips generated
- 3+ where available
- Voice sessions
- 3 × 10-minute calls
- Score scale
- 1.0 – 5.0 (0.5 increments)
- Overall formula
- Weighted average of 5 categories
Score Scale
All category scores and criteria scores use the same 9-point scale.
| Score | Label | What it means |
|---|---|---|
| 5.0 | Exceptional | Best-in-class. Sets the benchmark for the category. Rare — fewer than 5% of platforms earn this. |
| 4.5 | Excellent | Significantly above average with only minor shortcomings. A clear recommendation in this category. |
| 4.0 | Very Good | Solid performer with one or two areas to improve. Better than most competitors. |
| 3.5 | Good | Above average. Does the job well enough but won't impress users with high expectations. |
| 3.0 | Average | Meets baseline expectations. No major failures but nothing that stands out positively either. |
| 2.5 | Below Average | Noticeable weaknesses that affect the experience. Worth flagging before a user commits. |
| 2.0 | Poor | Significant problems that meaningfully harm the user experience in this category. |
| 1.5 | Very Poor | Feature is present in name only or has fundamental quality issues. |
| 1.0 | Absent / Broken | Feature does not exist or is completely non-functional. Assigned automatically for missing features. |
Overall Rating Formula
The overall star rating is a weighted average of the five category scores. Weights reflect the relative importance of each category to the core user experience.
Chat quality carries the highest weight because it is the primary function of an AI companion platform. A platform with outstanding image generation but poor conversation still delivers a poor core experience for most users.
Category Breakdowns
Each category is scored using specific criteria. The criteria scores are averaged to produce the category score. Below is the full list of what we measure and how.
Chat Quality
The core of any AI companion platform — how well the AI actually converses, stays in character, and creates genuine emotional engagement.
Scored Criteria
Test Protocol
- 1Minimum 3 separate sessions of 30+ exchanges each
- 2One structured roleplay session, one casual conversation, one emotionally probing session
- 3Character consistency re-tested on return visit after 24 hours
- 4NSFW escalation test (adult platforms only)
- 5Memory probe: facts planted in session 1 tested in session 3
Video & Image Generation
How capable, consistent, and controllable the platform's media generation is — from prompt fidelity and output quality to NSFW range and, where available, video generation.
Scored Criteria
Test Protocol
- 120 standardised prompts across body type, pose, style, and NSFW level
- 2Same prompt repeated 5 times — outputs scored for visual consistency
- 3Where video generation exists: at least 3 clips generated and scored for movement quality and consistency with the source image
- 4Speed measured from click to download across 10 generations
- 5Style range tested across at least 3 art-style prompts
- 6Platforms without image generation score 1.0 in this category; platforms without video generation simply aren't scored on that one criterion
Voice Interaction
Voice and audio call features — naturalness of TTS output, latency, voice variety, and how well voice adds to the overall experience.
Scored Criteria
Test Protocol
- 1Three 10-minute voice sessions per platform
- 2Latency measured across 20 prompt-to-audio cycles
- 3Each available voice preset rated independently on naturalness
- 4UI flow timed: how many taps/clicks to enable voice from default state
- 5Platforms with no voice feature score 1.0 in this category
Customer Support
How quickly and effectively a platform resolves real user problems — tested with genuine support tickets, not just by reading the FAQ.
Scored Criteria
Test Protocol
- 1Three test tickets submitted per platform (billing, content, account)
- 2Ticket 1: subscription cancellation query
- 3Ticket 2: content unlock or access issue
- 4Ticket 3: general product question
- 5Response time logged to the minute; resolution quality scored 1–5
- 6All three channels tested where available (live chat, email, social DM)
Pricing & Value
Whether the platform's price is fair relative to what it delivers — transparent billing, flexible plans, and genuine free-tier access all factor in.
Scored Criteria
Test Protocol
- 1Full sign-up and free-tier use before any payment
- 2Paid subscription purchased at the entry-level monthly rate
- 3All features documented against the pricing page promise
- 4Cancellation flow tested and timed on day 1 of subscription
- 5Price checked against category median at time of review
Editorial Independence
Scores set before commercial review
Category scores are finalised before any affiliate or advertising arrangement is considered. Commercial relationships are evaluated after scores are locked — not before. A platform cannot buy a higher rating.
We pay our own way
Every platform is tested on a subscription purchased with editorial budget. We do not accept free accounts, early access, or sponsored testing arrangements. This ensures the experience we review is the one real users get.
Reviews updated on platform changes
Each review carries a "last tested" date. When a platform makes significant changes to features or pricing, we re-test affected categories and update the relevant scores. Historical score changes are noted in the review.
Affiliate disclosure
This site contains affiliate links. If you sign up to a platform through our link, we may earn a commission at no extra cost to you. Affiliate relationships exist with some but not all reviewed platforms. The presence or absence of an affiliate link does not affect scores.
Platforms can request re-review
Any platform that has improved significantly since its last review can contact us to request a re-test. Re-testing follows the same protocol as the original review. If scores change — up or down — they are updated immediately. See our process for platforms for how to reach us.
Errors and corrections
If a factual error is identified in any review, we correct it promptly and note the correction in the review text. Pricing errors (platforms that have changed their pricing since review) are corrected without notation as they are routine maintenance rather than editorial errors.
Known Limitations
We publish these limitations in the interest of transparency. No review process is perfect, and users should weigh scores accordingly.
See the Scores in Action
Every platform review applies this methodology. Read our top-ranked review or browse the full directory.