AI POLICY

How Bluebell uses AI

Written for the person who has to sign this off — a parent, a head of department, a school data protection officer. It says what the model does, what leaves the service, and what we will not do with a child's work.

LAST UPDATED · 12 September 2026

You are talking to a machine

Every tutor reply on Bluebell is generated by a large language model. No human writes them, and no human reads them as they are produced. We say this on the page rather than in the small print because a student should never be uncertain about what they are talking to.

What leaves Bluebell, and what does not

The provider processes that data on our instructions, only in order to return a reply, and your work is not used to train any model. Our privacy notice names the provider, says where in the world your data is processed, and sets out the safeguards that cover it.

It gets things wrong

Language models state incorrect things confidently. This is not a bug we expect to fix, and no amount of care removes it. Bluebell is built to narrow the gap — the tutor works from a corpus written against the AQA, Edexcel and OCR specifications rather than from open-ended recall, and questions are checked before they are served — but a student should treat any answer as a good first draft and not as a mark scheme. Where an answer disagrees with a teacher or a textbook, the teacher and the textbook win.

What we do not use AI for

Human access

Conversations are stored on our server and are readable by the site administrator. That access exists to investigate faults and abuse, and to answer a deletion request — not to monitor students, and it is not routine. A parent or guardian may ask what is held about their child, and a school may ask about a student on its own code.

Honesty about schoolwork

Bluebell is built to explain rather than to hand over finished answers, and there is a Socratic mode in settings that makes the tutor work a student towards an answer instead of supplying one. It is off by default, and it is not a guarantee: this is a language model, and a determined student can get prose out of it. Text produced here must not be submitted as a student's own work. If your school has a policy on AI use, that policy governs — ours does not override it.

Telling us it went wrong

If the tutor states something incorrect, produces anything inappropriate, or behaves in a way that worries you, send it to raunaq@bluebellai.org with enough detail to find the conversation. Reports about a child's safety are looked at the same day.

See also the privacy notice and the terms.