Setting up an AI tutor for students

A tutor is scoped to a subject and, optionally, anchored to a unit so it answers in the context students are actually working in. It starts as a draft, invisible to students and unable to hold a conversation.

Your guidance shapes how it teaches. It sits beneath the platform's own rules rather than above them, so guidance asking it to hand out answers does not override the instruction not to.

Activation is what makes a tutor available. A draft tutor cannot be chatted to even by a student who knows its address.

Asked for finished work, it gives the next step and a question rather than the answer. That is a rule of the platform, not a preference you set.

It has no access to assessment records or other children. A tutor anchored to a unit knows the unit's title and inquiry, and nothing about the student it is talking to beyond the conversation.

A message that reads as distress never reaches the model. The reply is fixed, points the child to a trusted adult, and makes no promise of secrecy. The conversation is flagged and staff are told a flag exists — not what was written.

Every conversation is readable by staff, and opening one is recorded. That record exists because reading a child's conversation should be a considered act with a trail, not an ambient capability.