No affiliate rankings. No sponsored reviews. We buy our own accounts and use them for a month before writing a single word.
The five scoring axes
Every AI companion we review gets rated across five axes. These aren’t arbitrary — they’re the dimensions that matter most when you’re actually using these apps day-to-day, not just skimming a landing page.
Personality (25%)
How natural, warm, and consistent the companion feels across a long conversation. We measure this over 30 days of daily interaction — not just a first impression. Does the AI hold a coherent personality after 100+ messages? Does it adapt to your conversational style without losing its own voice? We track tone consistency, emotional range, and how often it produces generic filler responses.
Photo Quality (20%)
How good — and how consistent — the generated images are. We evaluate resolution, facial consistency across multiple generations, variety of poses and settings, and whether the AI maintains a recognizable appearance. We generate at least 20 images per platform during our testing window.
Memory (20%)
Whether it remembers specifics after a week, not just within a single chat. We test by sharing personal details in week one, then referencing them in weeks three and four without prompting. We track how many details are retained, how accurately they’re recalled, and whether the AI confuses details from different conversations. We put this axis to the test directly in our breakdown of which AI companions actually remember you after a month of use.
Value (20%)
What you actually get for the money versus the free tier. We compare feature-by-feature: what’s locked behind the paywall, what quality difference exists between free and paid tiers, and whether the premium features meaningfully improve the experience. A $20/month app that delivers twice the quality of a $15/month competitor earns a higher value score.
Price (15%)
The real monthly cost, including the features hidden behind the paywall. We document every pricing tier, track promotional pricing versus renewal rates, and calculate the true cost of the experience most users actually want. Hidden fees and misleading “starting at” prices get called out explicitly.
How we test
We purchase our own subscriptions to every platform we review — no free review copies, no affiliate-incentivized early access. Each app gets a minimum 30-day trial with daily usage. When a platform ships a major update, we re-test the affected scoring axes and update our reviews within two weeks.
Our current testing covers 12 platforms over 8 months of continuous use. Every comparison and review on this site draws from this same dataset — including head-to-head tests like our Candy AI vs Replika comparison and our full Candy AI review after three weeks of daily use and our Nastia vs Replika head-to-head, plus free-tier roundups like our best free NSFW AI girlfriend guide. The same Personality and Memory axes also shape our practical guides, such as how to deepen a Replika relationship and our editorial picks like the AI fashion models worth following in 2026 and our full Carla Kaas review, where we apply this same lens to a single AI persona. For our full data sources and pricing-audit process, see our testing methodology and our 2026 AI girlfriend app pricing study, where we priced six platforms against 18 sources.
How scores become rankings
Final scores are a weighted average of all five axes. We weight Personality highest (25%) because conversation quality is the single biggest factor in whether someone keeps using an AI companion past the first week. The remaining four axes split the balance based on how frequently readers tell us those factors influence their purchasing decisions.
Ties are broken by Value score, then by recency of the last major platform update. Apps that haven’t shipped meaningful updates in 90+ days get flagged with a staleness warning.
What we don’t do
No paid placements. No affiliate-driven rankings. No sponsored reviews. Our scores are never adjusted based on commercial relationships. If an app scores poorly, we say so — and we explain exactly why with the data to back it up.