A head of department should settle five things before the next deadline: which components are at stake, what AI use is allowed at each stage, what process evidence is kept, what a teacher does on suspicion and by when, and how the department supports the teacher who raises it. In England, the student's signature on the declaration is the practical cutoff for dealing with a doubt inside the school.
Ofqual's briefing pack on AI and coursework mentions heads of department once, and gives them nothing to do.
That is the gap. The pack, published for England on March 9, 2026, speaks to senior leaders: find out which qualifications carry coursework, brief your staff, tighten the policy. The work itself happens somewhere else, in a department, on a Tuesday, when a teacher reads a draft that does not sound like the student who wrote it. A head of department who wants to be ready should write five lines before the next deadline. I call them the Five-Line Statement, and the rest of this piece takes them one at a time.
Why does AI and coursework land on the head of department?
Because the school's policy says what is forbidden, and only the department decides what happens when a teacher has a doubt.
Look at where the official documents point. Ofqual's senior leadership team briefing pack tells leaders to ask the exams officer which qualifications include non-exam assessment, extended writing or on-screen coursework, so that "the right conversations are happening with relevant heads of department." The Joint Council for Qualifications (JCQ), the body that sets the malpractice rules for GCSEs and A levels, tells schools to make sure "teachers and Heads of Department are clear about their responsibility to only authenticate and submit work." And its guidance is plain that teachers "must be assured the work they accept for assessment and mark is the student's own work."
Assured is a demanding word. Nobody can be assured by a policy document. A teacher is assured by what they have seen: a draft in week three, a paragraph written in the room, a student who can explain why the second source matters. Those habits belong to the department, and so does the first response when they fail.
The Five-Line Statement is how I suggest a head of department settles AI and coursework: five lines, written before the next deadline, that say what is at stake, what AI use is allowed at each stage, what process evidence is kept, what a teacher does on suspicion and by when, and how the department supports the teacher who raises it. It is my suggested way of thinking, not research and not regulatory guidance.
What do the misuse figures actually tell you?
They tell you what was caught, which is a smaller thing than what happened.
Start with the numbers leaders quote. Ofqual's briefing pack says that in 2025, penalties included 1,125 cases in which students lost a whole GCSE or A level, and nearly 2,000 in which marks were deducted. The pack does not name its source. The figures match Table 3 of Ofqual's malpractice statistics for the summer 2025 exam series, published December 11, 2025, and there they cover every kind of student malpractice: 1,125 losses of aggregation or certification opportunity and 1,955 losses of marks. They are not AI figures.
The AI-specific number is smaller. The same release reports that plagiarism involving AI misuse accounted for 75.0 percent of student plagiarism cases in summer 2025, which is 100 cases, up from 85 in 2024. That is 2.0 percent of all student malpractice cases.
A hundred cases is not a measure of how many students used AI. It is a measure of how many were caught, reported and sanctioned. The Chief Regulator, Sir Ian Bauckham, wrote to exam board chief executives on March 2, 2026, and as I read the letter, it says detected cases are relatively low while the concern about the real extent of misuse is significant. He asked the boards for a more rigorous approach to candidate and teacher authentication of work.
My reading: do not plan around the detection figure, which is small and tells you little. Plan around the process, which you control.
What goes in the five lines?
Each line answers a question you would otherwise answer alone, under pressure, in the week of a deadline.
Line 1: What is at stake?
Name every component in your subject that counts toward a grade and is not a sealed exam. This is the pack's own first step, and it takes one conversation with the exams officer. The diagnostic question: could a teacher new to your department list those components without asking you?
Line 2: What AI use is allowed, stage by stage?
Planning, drafting and final submission each get their own answer, because "no AI" is a rule nobody can see being kept. JCQ's guidance sets the floor. When a student uses an AI tool as a source, the acknowledgment must show the name of the tool and the date the content was generated, and the student must keep a copy of the questions and the generated content and submit it with the work. JCQ also treats copying or paraphrasing AI-generated content, and failing to acknowledge AI use, as malpractice.
Your subject and your awarding body decide what is permitted. Your line should say it in words a Year 11 student can repeat. The diagnostic question: could a student tell you, without looking, which of the three stages allows an AI tool and what they must keep?
Line 3: What process evidence do you keep?
Keep what a conversation can later check. JCQ suggests examining intermediate stages of the work, allocating time for sufficient portions to be completed in class under direct supervision, and considering a short verbal discussion with the student about their work. In the language I use for assessment, that is the Three Ps: the Product is the finished piece, the Process is the drafts and any AI log, and the Performance is the moment the student explains it live.
A teacher who saw a draft in week three and heard the explanation in week six has an informed view. A teacher holding only the final file has a hunch. The diagnostic question: for each student, what would you have in hand if a doubt arose tomorrow?
Line 4: What does a teacher do on suspicion, and by when?
The first move is a conversation, and the deadline is the student's signature on the declaration. Ofqual's pack says a teacher who suspects AI-generated content should talk to the student first, asking them to explain their thinking and describe their process, and follow the school's malpractice procedures only if concerns remain. JCQ's guidance then draws a line in time. Where concerns arise before the student has signed the declaration, the school deals with them directly and does not report to the awarding body. After the declaration, the case must be reported.
So the signature is the real cutoff. The cheapest doubt to settle is the one raised before it. Write the date by which a teacher should come to you, name who they tell next, and confirm with your exams officer how it applies to each qualification, because the exams officer and the awarding body have the last word. If you want the process once an accusation has been made, the article on what a school should do when a student is accused of using AI covers it.
Line 5: How does the department support the teacher who raises a doubt?
In writing, with time attached. Ofqual's pack says teachers need clarity on what is permitted, confidence that they will be supported, time to review coursework carefully and talk with students, and a clear process for raising concerns. Its message to teachers is blunt: "You know your students. If something doesn't sound like them, question it. We will support you."
A department that makes the teacher who raised the doubt feel like the problem stops hearing about doubts. Say who sits in on the conversation if the teacher wants company, and where the marking time comes from. The diagnostic question: would your newest teacher raise a doubt, knowing exactly what happens next?
What does the statement look like written out?
It fits on one page and reads like house rules, not policy.
This example is hypothetical. A history department has a Year 13 independent investigation due in February, with twenty students and three supervising teachers.
- Line 1. The investigation is the only component at stake. Three teachers mark it; the exams officer holds the declaration dates.
- Line 2. AI may be used to suggest questions to ask of a source. It may not draft any sentence that appears in the work. Any use is acknowledged with the tool, the date and the prompts kept.
- Line 3. A dated draft in November. Twenty minutes of supervised writing in December from the student's own notes. A five-minute conversation about the argument in January.
- Line 4. Any doubt comes to the head of department by the Friday before declarations are signed, after the teacher has talked to the student.
- Line 5. One marking afternoon is released in the final week, and the head of department joins any conversation the teacher asks them to.
Your rules will differ, and the awarding body's rules for your subject come first. The shape is the point: a teacher can read it in two minutes and act on it.
What I see in practice
The pattern I see most often is a school with a clear policy and a department that has never read it.
Across the schools and leadership teams I work with, the policy says AI misuse is malpractice and the department says nothing. The teacher marking at ten at night wants something narrower than a policy. May I ask where this paragraph came from? What happens if the answer is bad? Who do I tell? As a former secondary teacher and assistant headteacher, I know how lonely that hour is when the answers are not written down anywhere.
The second mistake is reaching for a detector first. JCQ's own guidance says detection tools vary in accuracy, that they give lower scores for AI-generated content that students have since amended, and that they can be a useful part of the evidence a school considers. Part of the evidence, then, and not the verdict. I would start with the conversation, as Ofqual's pack does, and I explain the reasons in do AI detectors actually work. The wider design question, what to ask of students when AI can produce a competent draft, sits in what happens to assessment when AI can produce excellent work.
What does this not settle?
It settles a department's habits and leaves the rules to the awarding bodies.
Everything above applies to England and to JCQ-regulated GCSEs and A levels. Vocational qualifications, the International Baccalaureate, Advanced Placement and other systems have their own rules, and if you lead a department elsewhere, your awarding body's malpractice guidance replaces JCQ's in lines 2 and 4. The Ofqual pack is addressed to senior leaders, so what it implies for heads of department is my interpretation.
You might object that this is the exams officer's job. The exams officer knows which components exist and when declarations fall due. Only the department knows what the work looks like in week three. Both are needed, and line 1 is where they meet.
And you might ask whether a hundred detected cases justifies the effort. Ofqual does not treat low detection as low use, and the effort is one page and one meeting.
Where to start this week
Ask the exams officer for the list of components, then write line 4 first: the date and the name. Lines 2, 3 and 5 follow once you can see the calendar.
If your department heads are working through these questions, this is the kind of work I support through AI training for schools and leadership sessions. I also write about it in my newsletter.
Sources and further reading
- Ofqual, "AI and coursework integrity: senior leadership team briefing pack," GOV.UK, March 9, 2026: gov.uk
- Ofqual, "Malpractice in GCSE, AS and A level: summer 2025 exam series," GOV.UK, December 11, 2025: gov.uk
- Joint Council for Qualifications, "AI Use in Assessments: Your role in protecting the integrity of qualifications," effective April 30, 2025: jcq.org.uk
- Sir Ian Bauckham, Chief Regulator, Ofqual, letter to exam board chief executives, March 2, 2026: gov.uk (PDF)
Dan Fitzpatrick is the founder of The AI Educator, a former secondary teacher and assistant headteacher, and a Forbes contributor. More about Dan.
