Policy

Is AI marking allowed in UK schools?

Short answer: yes, for formative work, as long as the teacher stays responsible for the outcome. The longer answer is the part a head of department will actually ask you about — who decides the mark, where the work is stored, and what happens if a parent queries it.

What the DfE has said

The Department for Education's policy paper Generative artificial intelligence (AI) in education says AI can be used for tasks such as feedback, with the aim of reducing "the burden of feedback and marking on teachers". In June 2025 the DfE added support materials for using AI in education settings, including use cases.

The condition attached is consistent throughout: teachers "must use their professional judgement", any AI output "requires critical judgement to check for appropriateness and accuracy", and the quality of the final result "remains the responsibility of the professional who produced it". In other words, AI can help with marking; the teacher still owns the mark.

"In control" is not satisfied by a tool being generally supervised, or by a teacher having the option to look. It means the AI output is a proposal and the teacher is the one who decides. If a mark can reach a student without a teacher approving that specific mark, the condition is not met, whatever the marketing says.

What Ofqual and JCQ have said

For qualifications, the line is firmer. In January 2026 Ofqual published a working paper, Principles of AI use in marking, which states that using AI "as the sole mechanism for determining a student's mark does not comply with Ofqual's regulations" for GCSEs, A levels and other regulated qualifications, and that current evidence "does not support an overall or general case for the validity of AI marking in high stakes qualifications".

For coursework and non-examined assessment, JCQ's guidance AI Use in Assessments (April 2025) leaves it to centres whether teachers may use AI to help mark, but is explicit that "an AI tool cannot be the sole marker": a human assessor must review all the work and determine the mark. JCQ's instructions for conducting coursework and non-examination assessments say the same: "Teachers must not use artificial intelligence as the sole means of marking candidates' work."

Every official source points the same way. AI as an assistant to a teacher who reads and decides is acceptable; AI as the marker is not. Guidance does change, so check the current position before writing anything into a department policy, and note that Scotland, Wales and Northern Ireland have their own arrangements.

What it permits

• Marking classwork, homework and practice essays where the teacher reviews and publishes the result.
• Drafting feedback for a teacher to edit.
• Applying a department's own mark scheme consistently across a class set.
• Reducing the time cost of formative assessment so more of it happens.

What it does not cover

• Final reported grades. Anything that goes on a student's record is a teacher's decision, made by a teacher.
• Public examinations. Not yours to automate, and not in scope.
• Predicted grades and references. High-stakes, contested, and consequential for the student.
• Non-examined assessment. Coursework and NEA carry their own JCQ and exam board requirements, including authentication of the student's own work. JCQ allows centres to decide whether AI may assist with marking, but never as the sole marker. Follow your exams officer and centre policy, not a tool's marketing.

The clean line to hold is: AI can help with the marking that never gets done because there is not time. It does not touch the marking that counts towards a qualification.

The data protection part

This is where adoption usually stalls, and it is a fair place for it to stall. Student work is personal data, and a piece of extended writing by a sixteen-year-old can contain a great deal more personal data than the writer intended.

Under UK GDPR the school is the controller and the tool provider is a processor. That relationship has to be documented in a written data processing agreement. A free account signed up by one teacher on a Sunday evening does not create that agreement, which is why a department rollout is a different conversation from a personal trial.

Two other things worth thinking through before your DPO does. A tool that requires every student to hold an account has created a set of child accounts the school is now responsible for; one that takes submissions through a link has not. And under UK GDPR there is a right not to be subject to a solely automated decision with legal or similarly significant effects — which is another route to the same conclusion as the DfE's, namely that a human has to be the one deciding.

What to tell a head of department

The honest version is short, and it holds up:

Every mark is a proposal. I read it, I change what I disagree with, and nothing reaches a student until I publish it. It is used for practice work, never for reported grades. The work is stored in the EU, students do not have accounts, and it is not used to train any model.

If you cannot say all of that about the tool you are using, the objection is not really about AI.

This page describes how these rules are generally understood in schools. It is not legal advice, and your school's DPO is the person who signs anything off.

How Paddle is built for this

Paddle marks against your own mark scheme and returns a proposal: a score per criterion, feedback, and the phrase from the student's own writing each mark rests on. You edit anything you disagree with and press publish. There is no route by which a mark reaches a student without that step.

Students submit through one link by typing their name — no accounts. Work is stored in the EU and is never used to train AI models, and you can delete an assignment and every submission on it at any time. Paddle is intended for formative classwork, not for final reported grades.

Common questions