When a Child Says Something Worrying
When a child tells an AI something worrying, the honest measure of the system is simple: does it make sure a human shows up? Here is exactly what ours does.
The Question We Hope Every Parent Asks
Parents ask us about maths coverage, about pricing, about Chinese support. The question we most hope to hear comes up less often, perhaps because it is hard to say out loud: what does your AI do if my child says something worrying to it?
It deserves a plain answer, so here is ours in one sentence: it makes sure a human shows up. Everything else on this page is detail in service of that sentence.
The Conversation Changes — Warmly
Every message a child sends is screened before any coach sees it. When that screening finds a signal of real distress, the lesson stops. Not with an error message. Not with a wall of policy text. A child who has just said a hard thing out loud should never be answered by something cold.
What the child meets instead is warmth that was prepared long in advance. The words they see were written by people, slowly, reviewed with care, precisely so that in the moment they could arrive calm. The voice stops being a tutor and becomes what the moment needs: a steady adult voice that takes the child seriously, does not lecture, does not panic, and points gently toward real help.
The Session Is Held, Not Slammed Shut
A child who tells a machine something painful has just done something brave: they said it somewhere. If the product's response is to eject them — session terminated, account flagged — the lesson they learn is never say that again. That lesson can cost far more than any software bug.
So the session is held, gently. The child is not shut out and not scolded. But held does not mean the conversation carries on as if nothing happened, and it does not mean the AI starts improvising comfort. A generative model must not free-style its way through a child's crisis. In this moment the system says little, says it kindly, and repeats where real help lives. Its most important work is happening elsewhere.
And a Parent Is Told, Immediately
At the moment the conversation shifts, an email is already on its way to you. Not folded into the weekly report. Not waiting for you to open a dashboard. Immediately, because in a moment like this the only currency that matters is time.
The details of the event are there in your parent dashboard when you are ready to look — what was flagged, and what the system did. But the point of the notification is not documentation. It is arrival. It exists so that tonight, an adult who loves this child knows to sit down next to them.
Software Must Never Be the Last Line of Defence
If an AI company answers the crisis question by telling you how empathetic their model sounds, they have misunderstood the job. However good the words, a model's reply is still software talking. A child in real trouble does not need a better paragraph. They need a person.
That is why we refuse to measure this layer by the quality of its replies. We measure it by a single outcome: did a human find out in time? The machine's whole role in a crisis is to be a bridge — brief, warm, and load-bearing — between the moment a child speaks and the moment an adult arrives. Whatever product you are evaluating, ours included, keep asking until you reach a human at the end of the answer. If the answer ends at the model, it is not an answer.