Trial protocol
The three-session weekend trial for a free AI girlfriend
One long first chat measures novelty, and novelty is the one thing every companion app is good at. Three short sessions across a weekend measure the thing that actually matters: what is still standing when you come back.
- Three sessions, under twenty minutes each
- Markers planted on Friday, checked on Saturday
- Money decision only on Sunday
- A written line for each session
Sponsored waypoint. Candy AI is the labeled sponsor of this fieldbook. Treat it as one candidate on your route and run the same trial on it.
Sponsored: open Candy AI →Friday: the scout
Open the product on your chosen route and get to one real exchange. Before closing, plant two markers the service did not ask for: a name you want used, and one small invented detail, like a fictional dog called Biscuit. Ask one question that requires the reply to use the detail, so you know it registered.
Then leave. Ending the session while it is still enjoyable is the entire point of the scout. Write one line: how many screens stood between you and the first reply, and what each one demanded.
Saturday: the return
Come back on the same device through the same door. Do not re-introduce yourself. Ask the companion what your dog is called. A pass is the marker surviving overnight without prompting. A partial is a companion that keeps the tone but lost the facts. A blank slate on the guest route is expected and is data, not failure; a blank slate on an account is a real continuity result.
Spend the rest of the session on one boundary: state a tone limit and see whether three exchanges later it still holds. Log the return in one line: markers kept or lost, boundary held or ignored, and any new screen that appeared between you and the chat.
Sunday: limits and exits
The last session is deliberately unromantic. Find the edge of the free layer on your route: the message counter, the credit balance, the model selector, or the feature buttons that open an upgrade panel. Note what the upgrade screen names as unlocked, without buying anything.
Then locate two controls while you still feel nothing urgent: where an account or its data can be deleted, and where a subscription would be cancelled if you ever started one. If either control cannot be found in a few minutes of honest looking, that absence belongs in the log in capital letters.
The trial log
| Session | Question it answers | One-line record |
|---|---|---|
| Friday scout | What does entry actually cost? | Screens crossed and data requested |
| Saturday return | Does anything persist? | Markers, boundary, new walls |
| Sunday audit | What ends the free layer? | Limit found and exit controls located |
Reading your own log
Keep the product only if all three lines read calmly: entry matched the route, something you cared about persisted or you knowingly chose disposability, and you can name both the next limit and the way out. Any purchase decision made during a warm scene on Friday would have been made by the scene, not by you. The weekend structure exists to let the decision cool down twice before money enters it.
Trailhead questions
Why three sessions instead of one thorough test?
Because persistence and paywall placement only reveal themselves across separate visits. A single session, however long, cannot observe an overnight memory or a returning-user prompt.
What if the app is gone or changed by Saturday?
Log it and move on. A free tier that changes within a weekend has answered the continuity question by itself.