Is AI Marking Allowed in UK Schools? What the DfE's Guidance Actually Says

Why This Question Keeps Coming Up

Every head of department who looks seriously at an AI marking tool eventually asks some version of the same question: is this actually allowed? Not whether it works, but whether using it could land them in trouble with their SLT, with Ofsted, or with an exam board. It's a reasonable thing to want answered before you put a department's marking through it. It's also a question the government has already answered, in public guidance that's easy to find and short enough to read in ten minutes, which makes it strange how rarely anyone selling AI marking software actually quotes it.

What the DfE's Guidance Actually Says

The Department for Education publishes formal guidance on GOV.UK, Generative artificial intelligence (AI) in education, which applies to schools and colleges in England. Alongside it sits a plainer-language explainer DfE published on its own Education Hub blog, Artificial intelligence in schools: everything you need to know. Between the two, the position is consistent: teachers can use generative AI for tasks including marking work and giving feedback, but as the Education Hub post puts it, "the final responsibility always rests with them and their school or college." Staff are expected to check that anything AI produces is accurate and appropriate before a pupil sees it. The formal guidance puts the same idea more bluntly: "Generative AI tools can make certain written tasks quicker and easier, but it cannot replace the judgement and deep subject knowledge of a human expert."

Reporting on DfE's own training materials for teachers was more specific about where that applies in practice. According to that reporting, the materials pitch AI marking at "low-stakes" work: homework and quizzes are the examples given, not anything that counts toward a certificated grade. The thread running through all of it is the same: AI can take a first pass, a teacher has to be the one who actually decides.

Where the Line Gets Strict

That permissiveness has a hard edge, and it isn't DfE's to draw. Ofqual, the regulator that actually governs qualifications like GCSEs and A-levels, wrote to every awarding organisation back in September 2023 to confirm that using AI as the sole marker of a learner's work doesn't comply with its regulations. The reasoning is specific: Ofqual's own definition of an "Assessor" requires a person who is able to make a judgement, and an AI system marking alone doesn't meet that bar. Ofqual's published approach to AI in the qualifications sector hasn't moved from that position since: AI can complement and quality-assure human marking, but it isn't allowed to replace the human making the actual decision on anything that results in a certificated grade. The Joint Council for Qualifications' own guidance on AI in assessment, which DfE's formal guidance points to directly, exists for the same reason, protecting the integrity of the exam itself from any tool, AI or otherwise, that could compromise it.

Put the two pieces of guidance together and the shape of the rule is simple enough to state in one sentence: a teacher using AI to help mark homework, classwork or an internally set paper, and checking every mark before it counts, is squarely inside what the government has said is fine. A process where AI is deciding a mark with nobody checking it, on anything that reaches an exam board, is not, and no AI marking product changes that by being better at marking.

Where GradeDrive Sits Against That Line

Worth being precise here rather than stretching the guidance to fit. DfE's own named examples are homework and quizzes. Most of what actually gets marked on GradeDrive is heavier than that: mock exams, past papers, and assessments a department has set itself, not homework in the literal sense of DfE's examples. That's a real difference, and this post isn't going to paper over it by implying the guidance was written with mock exams specifically in mind, because it wasn't.

What does carry straight across, because it's the part of the guidance that actually matters, is the requirement that a teacher checks every mark before it counts. That isn't a setting a department can switch off in GradeDrive, it's how the product works for every submission: GradeDrive reads a student's answer against the mark scheme a teacher uploads, suggests a mark with its reasoning shown alongside it, and nothing is recorded until a teacher has seen that mark next to the student's actual answer and the criterion it matched, and has accepted or changed it. See the full workflow.

Two things GradeDrive deliberately stays away from. It never touches the actual certificated exam: that paper is marked by the exam board's own examiners, under Ofqual's rules, and nothing about GradeDrive is involved in that process at any point. And nobody here is claiming the DfE guidance "endorses" GradeDrive, or any other named AI marking tool. The guidance describes a responsible way of using AI to mark; GradeDrive is one way a teacher could work inside it, provided the teacher actually does the checking the guidance asks for, which is also the one step GradeDrive doesn't let anyone skip.

What "A Teacher Checks It" Looks Like, Not Just as a Policy Line

It's worth being concrete about this rather than leaving it as a compliance abstraction. Across summer 2026, GradeDrive processed 216,000 marks over 2,300 submissions, and 99.2% of those marks survived a teacher's review exactly as suggested. The other 0.8%, around 1,750 marks, is that review step doing precisely what the guidance asks of it: catching the cases where a teacher's judgement differed from the AI's first pass, before anything was recorded. The full numbers are here, and the broader question of where AI marking is strongest and where it needs more of a teacher's attention is covered in How Accurate Is AI Exam Marking?

Frequently Asked Questions

Is it legal for a UK teacher to use AI to mark homework?

Yes. DfE's own guidance and training materials for teachers describe this as an acceptable use of AI, on the condition that a teacher checks the output before it reaches a pupil. That condition applies to every example DfE gives.

Can AI mark a GCSE or A-level mock exam?

DfE's published examples are homework and quizzes specifically, not mock exams, so this post isn't going to claim the guidance was written with mocks in mind. The underlying principle, that a teacher reviews every mark before it counts, applies the same way regardless of the paper, and it's also how GradeDrive is built to work for every submission it marks.

Can AI mark the actual GCSE or A-level exam that counts toward a certificate?

No. Ofqual confirmed to every awarding organisation in September 2023 that AI can't be the sole marker of anything that determines a certificated grade, and that position hasn't changed. Certificated exams are marked by the exam board's own examiners under Ofqual's rules. GradeDrive has no role in that process at any point.

Does the DfE's guidance endorse GradeDrive or any other AI marking tool?

No, and this post isn't claiming otherwise. The guidance describes a responsible way of using AI; it doesn't name or endorse any product, GradeDrive included.

Where can I read the guidance myself?

DfE's formal guidance is on GOV.UK under "Generative artificial intelligence (AI) in education," with a plainer version on DfE's Education Hub blog. Ofqual publishes its own approach to AI in the qualifications sector separately, also on GOV.UK.

Internal Links

For what "a teacher checks every mark" looks like in practice, see AI Marking for GCSE and A Level. For the summer's full usage figures, see AI Marking in Summer 2026. For where AI marking is strongest and where it still needs a teacher's closer attention, see How Accurate Is AI Exam Marking?. For how AI marking compares to marking by hand, see AI Marking vs Human Marking. For the questions a department should ask before buying any AI marking tool, see AI Exam Marking Software: 5 Questions Every School Asks Before Buying.