AI Feedback in Schools: What Leaders Need to Know Before Adopting

AI feedback is not the same as AI marking

Many school leaders conflate AI feedback with AI marking — and understandably resist both. But they are different things. AI marking replaces professional judgement and issues a verdict. AI feedback is bounded, advisory guidance on a student's own draft, generated under controls the school sets and can override at any time.

The distinction matters because it determines governance posture. If the school retains control of how AI is used and what students see — setting the controls, reviewing where it chooses to, and with concerning content always held for staff — AI becomes a productivity tool rather than an autonomous decision-maker. That's a much easier conversation with governors and parents.

The school-control question is non-negotiable

Any AI tool used in a school context must pass one test: does the school stay in control of how AI is used and what students see — able to see, review, edit and override it, and to require staff sign-off where the stakes demand it? If the answer is no, the tool is not appropriate for school use.

This means real controls the school owns — not a black box that decides for itself, but a genuine control point where the school sets the policy, anything flagged as a concern is held for a member of staff, and nothing about a child's writing is reduced to an autonomous verdict.

Data safety is about specifics, not reassurance

When evaluating AI tools, ask specific questions: Where is data processed? Is it used for model training? Who can access student content? What happens when the contract ends?

Generic reassurance ('we take data seriously') is not sufficient. Look for specific commitments: UK data residency, no model training on student data, documented retention and deletion policies, and a Data Processing Agreement.

• Where is student data stored and processed? (Look for UK/EU data centres.)
• Is student data used to train or improve AI models? (The answer should be no.)
• What happens to data when the school's contract ends? (There should be a defined deletion process.)
• Is there a DPA template available? (This should be standard for school procurement.)

Start with a pilot, not a policy debate

The fastest way to resolve internal questions about AI in your school is to run a controlled pilot with real students. Abstract policy discussions about AI tend to go in circles — practical experience with a specific tool gives your team concrete evidence to evaluate.

A good pilot has clear success criteria (completion rates, workload impact, feedback quality), defined scope (one cohort), and a review point where leadership can decide whether to scale based on your school's data.