Methodology
How I test AI girlfriend and AI companion apps
I'm Sam. I pay for AI girlfriend and AI companion apps with my own card, then test the parts their marketing and most review sites do not prove: long-term memory, character consistency, image and video quality, filters, real media costs, privacy controls, and cancellation. Every report labeled hands-on tested follows the same protocol. Research-stage directory and ranking data is labeled separately, and every measured number carries the test date.
The protocol
1. I subscribe like a normal user. Card on file, no press accounts, no partner onboarding. Prices quoted are the prices actually charged at checkout. When they differ from the advertised price, I say so.
2. I measure the free tier. Message caps, locked features, and where the paywall actually lands, tested by using the app, not by reading the pricing page.
3. I test memory across sessions. A companion does not get credit for repeating something from the previous message. I plant facts, return in later sessions, vary the wording, and record whether the app recalls the fact, invents a substitute, or loses it completely.
4. I test character and media consistency. I rerun controlled scenarios and prompts, count refusals and failed generations, and keep the bad outputs. Image and video scores include face drift, instruction misses, visible artifacts, retries, and cost per usable result.
Privacy and cancellation
5. I count the cancel steps. From account page to confirmed cancellation, every click and every retention screen is counted and screenshotted. Guilt-trip screens and hidden buttons get named.
6. I read the privacy policy so you don't have to. Whether your chats train their models, whether you can delete your data, what their age checks consist of, and who handles payment, quoted with section numbers.
7. I date everything. AI companion apps change prices, models, limits, and policies quickly. Every published test figure carries its measurement date and is scheduled for rechecking.
Tested vs researched
Every article on this site carries one of two labels. Tested hands-on means the full protocol above: my own subscription, my own screenshots, dated numbers. Researched means the full protocol has not run yet; the article is built from primary sources instead: official pricing pages, privacy policies read line by line, breach disclosures, and documented community reports. Researched articles never contain invented test numbers, and they name their sources. When a researched app graduates to a full test, the label changes and the date resets.
What I don't do
No sponsored placements in rankings. No review copy approval for vendors. No undisclosed affiliate links. The disclosure page explains exactly how this site makes money, and rankings are locked before partnerships are signed.
See the evidence
Start with the paid AI girlfriend app comparison, the AI companion pricing tracker, or the sourced AI companion statistics. Individual reviews link their test date, plan, cost, evidence limits, and the exact parts I tested.
Corrections
If I got a number wrong, I fix the page, note the correction, and re-date it. Corrections and retests are part of the protocol, not an embarrassment.