AI Trust Centre
How Homework Assist's AI actually works — plainly, and accurately.
Last updated
Why this page exists
Homework Assist uses AI to help students work through a problem step by step. Parents and students deserve a plain, honest description of what that actually means — not marketing language. Everything on this page describes real, already-built behavior. Nothing here is aspirational.
What happens on every AI call
A daily limit, always
Every student has the same platform-wide daily cap on AI-assisted Homework Assist help, purely as cost/abuse insurance.
Every message is screened
What a student submits is checked before it reaches the AI, and what the AI writes back is checked again before it reaches the student.
Answers follow a fixed shape
The AI never returns free-form text. Every response is validated against a strict, predefined structure — a malformed or unexpected response is rejected, not shown.
Every call is logged
Every AI request — what it was for, when it happened, whether it succeeded — is recorded for oversight and cost control.
Fails closed, not open
If a safety check itself breaks, the request is blocked, not waved through. A guardrail failing never means a guardrail being skipped.
Who decides what
The AI's role is strictly limited to reasoning through a step and phrasing a message. It never decides policy, never judges a student, and never acts on anyone's behalf. A human — a tutor, a teacher, a parent, or the student themselves — always remains the one who decides what happens next.
Hints never skip ahead
A hint request always advances by exactly one level of a fixed ladder. There is no way to jump straight to a stronger hint or the answer itself — and no level ever states the final answer outright.
A tutor decides, never the AI
When a student shows a real pattern of struggling, a human tutor is shown the evidence and decides whether and how to follow up. The AI is never asked to judge whether a student "deserves" help, and it never acts on a student's behalf.
The system decides the facts, the AI only phrases them
Briefings shown to a tutor are built entirely from real, already-recorded data — questions attempted, hints used, topics struggled with. The AI may turn those facts into readable prose, but it never adds a new fact, and never offers an opinion of its own about the student.
Nothing the AI says about a problem is trusted as fact
Where a teacher provides their own marking criteria, the AI's role is limited to phrasing — the teacher's own criteria are always enforced exactly as written, never replaced by anything the AI generates.
What's never invented
Every number and claim shown anywhere in Homework Assist or the Learning Passport traces back to something a student genuinely did, or something a parent or teacher genuinely entered. None of the following is ever guessed, predicted, or fabricated:
- A goal's progress percentage — always a real ratio between an actual measurement and a target the student set, never a guess.
- Whether a goal was achieved — only ever marked by the student themselves, never inferred.
- A topic mastery level — only ever built from questions a human reviewer has actually approved.
- A "struggling" or "declining" pattern — always traced back to real, recorded evidence; never a forecast, prediction, or invented risk score.
- A recommendation shown to a tutor — never goes beyond what the recorded evidence actually shows.
What data is used, and by whom
- Homework Assist only ever sees the specific question a student is working on — not their broader browsing or personal information.
- A parent can see real, recorded patterns about their own managed child's activity — never another family's.
- A tutor only ever sees evidence for a student they have an actual booking relationship with.
Questions about how this works?
This page will always describe what Reclass actually does, kept up to date as Homework Assist evolves. If anything here seems unclear, our Privacy Policy has more detail on how student data is handled generally.