Draft: not yet in force. This document still contains details marked [TO CONFIRM]. It is published for review and should not be relied on until those are filled in.

How drafts are made

The privacy policy says this in legal language. This says it properly. If you read one page before typing a child’s name into something, make it this one.

Last updated 29 July 2026

What happens when you press generate

You fill in the form: mostly dropdowns, with a little free text.
Your answers are combined with written guidance on teaching practice: strengths-based language, trauma-informed framing, restorative approaches, and rules about handling information about children.
That whole package is sent to Claude, the AI model made by Anthropic, and processed on their servers in the United States.
The draft is written back to your screen as it is generated.
It is saved in your browser so you can come back to it. It is not saved anywhere else.

What Claude is given

Everything you typed into the form, including the student’s first name and whatever you wrote about the situation. There is no way to generate a draft without sending it. That is what generating means.

Alongside your answers, two small things about you are included:

Your first name, so the draft can be signed off properly instead of leaving a placeholder.
Your state, so it uses the right terminology: Prep, Foundation, Kindergarten and Reception all mean the first year of school somewhere.

What Claude is never given

Drafts you have already written or saved.
Your email address or your school’s name.
Anything belonging to another teacher.
Anything you did not type into that form.

What Anthropic does with it

Under Anthropic’s published terms for API customers like three30, inputs and outputs are deleted from their systems within 30 days, and retained data is not used to train models without express permission. Content flagged by their automated safety systems may be kept for up to two years.

Anthropic offers no Australian processing region, so this happens overseas. Requests are pinned to United States infrastructure rather than being sent wherever there is capacity. It cannot be Australia, but it is at least somewhere known and fixed. You can read their retention policy and privacy policy directly rather than taking this page’s word for it.

Why first names only

Claude is instructed never to ask for, repeat or infer a full name, address, date of birth, diagnosis, or anything about a child’s family circumstances, and to reduce any such detail to an initial if it appears. It is also instructed not to invent incidents, quotes or history that you did not describe.

Those instructions are reliable, not guaranteed. The way to be certain a detail is not sent is not to type it. A first name is genuinely enough for every tool here.

If a child might be at risk

When something in what you have written suggests a child may be at risk of harm, three30will not write the draft. It returns a note pointing you to your school’s child-safety and mandatory reporting procedures instead.

That is deliberate and it is not a failure of the tool. Those situations need a person, your leadership or wellbeing team, not a drafted email.

It can be wrong

Claude writes fluently, which makes mistakes harder to spot rather than easier. It can misjudge tone, overstate what happened, soften something that needed to stay firm, or produce a sentence that reads beautifully and is not true of the child in front of you.

Read every draft properly before you use it. You know the child, the family and the history; it knows a form.

Where your drafts live

In your browser, on the device you wrote them on. Not on a server. There is no database table for drafts. That is a deliberate structural decision, not a setting.

Which also means there is no backup. Signing out deletes them. Clearing your browser data deletes them. Another teacher signing in on the same device deletes them. Copy anything you need to keep into your school’s systems.