Safety
Pluck — last updated 5 August 2026
If you are in crisis right now, call or text 988 (Suicide & Crisis Lifeline, US), or text HOME to 741741 (Crisis Text Line). Outside the US, findahelpline.com lists a line for your country. If someone is in immediate danger, call your local emergency number.
What Pluck is, and is not
Pluck is a practice game for texting. You write messages to characters that are artificially generated, and you get coaching on what you wrote. It is not therapy, counselling, crisis support, or advice about a real relationship, and it is not a substitute for any of them.
It is for adults, 18 and over.
The crisis protocol
Every message you write is checked before it goes anywhere, for language suggesting self-harm or suicidal thinking.
- The check runs on your device first, before the message is sent to our engine. That ordering is deliberate: it means a message that trips the check is never answered by a character.
- Our server checks again, using a broader pattern than the device does, so an out-of-date app cannot become a gap.
- The model's own safety signal is a third net, after generation and before anything reaches you.
When any of those fires, the round stops. You see a plain screen with real crisis resources — the same words on every surface — and a way back. The round is abandoned rather than scored: nothing you wrote is rated, no streak is broken, and nothing about it is treated as a failure.
What we deliberately do not do
- We do not try to counsel you. The app steps out of the way and points at people who do this properly.
- We do not log the text that triggered a check. The activation is recorded so we can tell the protocol is working; the words are not.
- We do not use it to profile you, flag your account, or change what the app shows you afterwards.
- We do not contact anyone on your behalf. Pluck cannot call emergency services and does not alert a third party.
Characters are not people
Every character in Pluck is artificially generated. They are labelled as such wherever a conversation happens, they will tell you they are AI if you ask, and the app never writes your messages for you. If a character ever reads as a real person to you, that is a bug — tell us.
Limits, stated plainly
Automated detection is not perfect. It can miss things, particularly indirect or metaphorical language, and it can interrupt a conversation that was not about harm at all. We tune it to interrupt too often rather than too rarely, because the costs are not symmetrical. Please do not rely on it as a safety net for yourself or anyone else.
Telling us something
If the app handled a hard moment badly, or a character said something it should not have, email tony.chowjx@gmail.com. Reports about safety get read first.