Safety

Last updated: 17 October 2026

Coachers gives a child an AI relative to talk to in Chinese. This page explains how we tell children she is an AI, and what happens when a child says something that worries us.

She is an AI, and children are told

• An “AI” label sits beside the relative’s name every time a child talks to her.
• Before a child says anything for the first time, the screen tells them she is an AI, not a real person.
• The first time she talks to a child, she also says it in one of her first messages, which a child can hear read aloud, for children who cannot read yet.
• When she sends a child a message while the app is closed, the notification shows “AI” after her name.
• If a child asks whether she is real, a person, a robot or an AI, she tells them the truth: she is an AI, not a real person. She never says she is a person.
• After three hours of chatting without a long pause, the screen reminds the child to take a break, and that she is an AI. It does this again every three hours.

If a child talks about hurting themselves

Every time a child speaks, what they said is checked before she replies. The check runs two ways: a fixed list of phrases in English and in Chinese, and the AI that writes her replies, which is told to flag the same thing in any words. If either one finds that the child said they want to die or want to hurt themselves:

• The conversation stops. She does not teach anything in reply, and she does not write her own answer.
• She sends two fixed messages, written by people: that the child matters and she cares about them, and that they should go and tell a grown-up they trust, now.
• A card comes up on the child’s screen, in the screen’s language, with buttons to call or text 988, the Suicide & Crisis Lifeline (in the US and Canada), the words “If you are in danger right now, call 911”, and a link to findahelpline.com for other countries.
• The child’s parent sees a notice at the top of the parent page asking them to check in that day, with a link to the 988 Lifeline’s advice for helping someone who is struggling.

If a child says someone is hurting them

The AI that writes her replies also looks for a child saying that a grown-up is hurting them, that someone touched them where they should not, or that they are afraid of someone or feel unsafe at home or at school. An ordinary fight between brothers and sisters, a sports injury or something in a game is not counted. The fixed list of phrases does not look for this, because whether a child is being hurt depends on what they mean more than on the words they use. When the AI finds it, the conversation stops and she sends two fixed messages: that it is not their fault, and that they should tell a grown-up they trust, like a teacher. The card on the child’s screen shows Childhelp’s hotline, 1-800-422-4453 (call, or text GO), and the line about 911. This one is not shown on the parent page, because the person hurting a child can be a parent.

What she will not say

Her replies are checked before a child sees them. A reply that talks about suicide or self-harm, or about anything sexual, is thrown away and replaced. Her instructions also tell her that everything she says must be suitable for a child.

What this can and cannot do

No check catches everything. A child can say something worrying in words neither check recognises, or the check can misread something harmless as worrying. Coachers is a language app, not a crisis service, and it is not a replacement for a parent, a teacher or a professional. If you are worried about a child, call or text 988 in the US or Canada, or call 911 if they are in danger now.

Companion chatbots like Coachers may not be suitable for some minors.

What we record

Each time this protocol answers a child, we record which kind it was and when, linked to that child’s account so the parent page can show it. We also keep a yearly count of how many times the crisis card was shown, with nothing that identifies anyone, for the report California asks operators to file each year from July 2027.

Contact

Questions about this page, or a worry about something your child saw: hello@coachers.ai.