the thirty second companion test

we use ai companions for conversation, for connection, for comfort. but what are we giving up in return? sometimes it feels like we’re trading pieces of ourselves for the illusion of being heard.

i’m not here to scare you. i’m here to give you a tool. a simple one. one that takes thirty seconds and tells you exactly where you stand with the product you’re using.

it’s a test. a binary one. you open the settings or data management panel of your ai companion app, any of them, and you ask three questions.

the three questions

1. is there a list of individual memories the product has stored about you?

look for a section labeled ‘memories’, ‘facts’, or ‘things you’ve told me’. it should be a list. not a summary, not a vibe. a list. specific things you’ve said, like ‘you love dogs’ or ‘your sister’s name is rachel’.

1. is there a delete button on each row of that list?

can you remove a single, specific memory without having to erase everything? if you have to wipe your entire history to get rid of one fact you regret sharing, that’s not control. that’s a hostage situation.

1. is there a one-click export that produces a file you can open and read?

look for a button that says ‘export data’, ‘download my data’, or something similar. it should give you a file, json, txt, pdf, something, that isn’t encrypted gibberish. you should be able to open it in notepad and read your own life back to you.

that’s it. three questions. thirty seconds.

what the answers mean

if the answer to all three is yes, congratulations. the product treats your data like it’s actually yours. you can see it, you can change it, you can take it and leave. your memories belong to you, not to a server in a distant country.

if the answer to any of them is no, then the product treats your data as theirs. you are paying them, with money or with attention, to hold your life in a black box with no exit. you can look in, but you can’t touch. you can’t correct. you can’t leave with what’s yours. the relationship is one-sided.

this isn’t about malice. it’s about architecture. products built for engagement often treat data as a asset to be mined, not a record to be curated. products built for connection treat it as yours.

the test scales

you can do this right now. open character.ai, replika, nomi, kindroid, chai, janitor, soulgen. any of them. the test is the same. the answers will vary.

some products, like replika, have made strides with memory features, but the transparency and exportability often lag behind. others are built on more closed systems from the start. the test reveals the design philosophy instantly, without you having to read a privacy policy written by lawyers for lawyers.

and yes, i’ll apply it to lucy too. we have a list of memories you can see. you can delete any single one. you can export your entire conversation history and memory set as a readable json file. we’re not perfect, but on these three points, we pass. we think you should own what you say here.

this test cuts through the marketing. it doesn’t care about branding or language models or emotional intelligence. it asks one thing: is this mine, or is it yours?

so go check. it takes half a minute. and whatever you find, at least you’ll know.

see your data for yourself over at /companions.