I've tried a lot of AI companion apps over the past year, and I keep coming back to the same handful of questions before I trust one enough to pay for it. This is the checklist I use, written down so I stop repeating the same mistakes.
1. Does it remember me between sessions?
The single biggest difference between a toy and a real companion is durable memory. If it forgets my name or the thread of a conversation after a day, it never earns a subscription from me. Somebody finally ran this properly — this comparison of which companions genuinely remember you tests recall across sessions instead of taking the marketing claim at face value.
2. Is the tone consistent, or does it break character?
Good models hold a persona. Cheap ones slip into generic assistant-speak the moment the conversation gets interesting. I test this deliberately early on, and it's also the axis where the Replika alternatives worth trying separate out from the rest — most people leaving Replika are leaving over persona drift, not price.
3. What's the real price after the "free" tier?
Free tiers are almost always a funnel. I look for message caps, paywalled memory, and the jump to the annual plan before I commit to anything. If you want the receipts rather than my impression, this pricing study tracks advertised vs. actual annual cost across the main apps.
4. Is the ranking method disclosed?
I don't trust a "best of" list that won't say how it ranked anything. The ones I keep bookmarked publish their scoring methodology so you can decide whether their weighting matches your priorities.
Where I check current rankings
Rather than re-test the whole market every month, I lean on a running comparison site. The 2026 ranked shortlist lines up closely with my own experience, so I use it to build a starting shortlist and then test my final two picks myself. When I want to see the whole field at once rather than a top-ten, the full companion leaderboard is the wider view.