Step one: pay like a real customer
Each test starts with a fresh signup. I build a companion from nothing and buy the tier a typical reader would choose. Free trials are fine for judging a welcome screen, but twenty messages will never show you whether the character remembers you by day twelve.
From then on it is a daily habit. My notes track three things: which details stuck, which ones the app invented, and how painful it was to get my data out or delete it. The renewal price goes in the log too.
The scorecard never changes
| Category | The question I am answering |
|---|---|
| Conversation | Is it still coherent in week two, or was one sharp reply a fluke? |
| Memory | Do small facts come back on their own? |
| Media | Do the pictures and the voice belong to one consistent character? |
| Price | What does the renewal cost, and what do tokens and caps add? |
| Controls | Can you export, delete, adjust content settings, and is there an age check? |
The final rating averages those five and rounds to a single decimal. Once set, it is frozen until the next monthly pass, no matter how exciting the latest update notes sound, and it is set before any company reaches out.
Money gets its own audit
List prices rarely tell the whole story. I put the first-month offer next to the renewal, check whether each image eats tokens on top of the plan, and judge whether the free tier is a usable product or just a preview. That work is written up in the real price of AI girlfriend apps.
Privacy is scored, not skimmed
People tell these apps personal things. So I read the privacy policy line by line, search it for anything about encryption, and try deleting and exporting my own account. Fuzzy promises such as "data may be shared with partners" are flagged in the review. Their infrastructure is invisible to me, and whenever I am inferring rather than confirming, the review says so.
Not for sale
Rankings cannot be bought, and paid placement offers get a polite no. The affiliate side is explained on how we make money; it has no effect on positions. If an app slips, its score falls at the next review. Spot something out of date? Write to the address in the footer.
Where outside testing stops
I can read their policies and push every button they give users, but I cannot audit their servers. When an answer is vague, that vagueness goes into the review as a result in its own right.
