Paid accounts, not demos
For each app I create a fresh account, build at least one companion from scratch, and pay for the plan most readers will actually use. A twenty-message free trial tells you how the onboarding looks. It does not tell you whether the thing still knows your name on day twelve.
I talk to it daily. I log what it remembered, what it invented, what the renewal price was, and how hard it was to export or delete anything.
The same five scores every time
| Score | What I am testing |
|---|---|
| Conversation | Coherence over weeks, not one clever reply |
| Memory | Unprompted recall of small details |
| Media | Images and voice that belong to the same character |
| Price | Renewal cost, tokens, fair-use caps |
| Controls | Export, delete, content settings, age gates |
The published rating is the average, rounded to one decimal. I lock it before anyone from the company emails us. A changelog does not move a score mid-cycle. It waits for the next monthly pass.
Hidden costs get their own pass
Sticker prices lie. I check intro versus renewal, whether images burn tokens on top of the subscription, and whether the free tier is a product or a trailer. The longer write-up is what these apps really cost.
Privacy is a score, not a footnote
These chats are intimate by design. I read the privacy policy, look for encryption language, and try the delete/export path. Vague “we may share with partners” copy gets flagged in the review. I cannot see their servers. When I am guessing, I say so.
What a score is not
It is not for sale. Affiliate links, explained in the disclosure, do not change a ranking slot. We decline pay-to-play. If an app gets worse, the number drops on the next pass. If you spot something stale, the footer address is the correction line.
What I cannot see
I am not on their servers. I can read the policy, try export/delete, and tell you when the language is fog. Unclear is a finding.
