AI in Exam-Prep Teaching: A Practical Guide to Assessment Integrity

AI can be useful in exam-prep teaching, but it changes the questions educators need to ask. The key question is not simply, “Is AI allowed?” It is: “Does this activity still give us a trustworthy view of what this learner understands and can produce on their own?”

That distinction matters most when teaching supports regulated qualifications in England. Ofqual’s published approach recognises both opportunities and risks from AI in the qualifications sector. Its regulatory objectives include fairness for students, validity of qualifications, assessment security, public confidence and innovation. While Ofqual regulates awarding organisations rather than independent tutors or course creators, those objectives provide a useful practical lens for anyone designing exam preparation.

This is an educational workflow, not legal or awarding-body advice. Always check the current rules, specifications and instructions issued by the relevant awarding organisation, school, college or examination centre before treating an activity as formal assessment preparation.

What Ofqual’s AI approach means for exam-prep educators

For an exam-prep provider, assessment integrity is about preserving a clear boundary between supported learning and evidence of independent performance. AI can widen access to explanations, generate alternative practice questions and help learners notice gaps in a draft. But it can also supply language, reasoning, calculations or answers that a learner cannot reproduce under exam conditions.

A sensible working principle is:

Use AI freely enough to support learning, but control or remove it whenever you need dependable evidence of independent exam readiness.

This principle prevents two unhelpful extremes. A total ban can deny learners useful opportunities to question, practise and reflect. Unrestricted use can make mock-exam scores, homework and portfolios difficult to interpret. The aim is not to make every activity AI-free. It is to label the conditions of each activity clearly and match them to its purpose.

Where AI can support learning without replacing student thinking

Helpful AI use in exam prep normally leaves the learner doing the intellectual work: recalling, selecting, applying, explaining, checking and improving. The tool may make practice more responsive, but it should not silently complete the performance being practised.

• Explanation from multiple angles: Ask for a simpler explanation, a worked non-assessment example or an analogy, then ask the learner to explain the idea back in their own words.
• Retrieval practice: Generate short questions from teacher-selected content. Require the learner to answer before seeing any feedback.
• Planning prompts: Use questions such as “What evidence would strengthen this argument?” rather than asking for a finished response.
• Error analysis: Give AI an anonymised, teacher-created misconception and ask it to produce diagnostic questions. The learner then identifies and corrects the error.
• Accessible rehearsal: Rephrase a teacher-approved explanation or produce a structured revision checklist. Check that the rephrasing has not altered subject meaning.

The practical safeguard is simple: design a response task after the AI interaction. For example, a learner might close the tool and write a three-sentence explanation, solve a parallel problem, annotate a source, or record a brief verbal justification. This converts passive consumption into observable learning.

AI uses that need extra caution

Risk rises when AI creates work that is likely to be mistaken for the learner’s own independent output. This includes full essays, completed calculations, generated coursework sections, answers to secure or unreleased materials, or polished rewrites that substantially change the learner’s original work.

Be particularly careful with timed mock exams, baseline diagnostics, progress checks used to make placement decisions, and any work that learners may later submit as evidence. If AI is available during these activities, the resulting score may measure a mix of learner knowledge, prompt-writing skill and tool output rather than exam readiness.

Extra caution does not necessarily mean automatic prohibition. It means deciding in advance what is permitted, communicating it, and recording enough information to interpret the result. If a learner used AI to brainstorm before a practice essay but wrote independently afterwards, that may be a valid learning activity. It should not, however, be reported in the same way as an unaided timed essay.

A practical AI-use policy for tutors and course creators

A short, plain-language policy is more likely to be used than a long generic document. Build it around activity labels that learners can recognise immediately.

1. Green: AI-supported learning. AI may be used for explanation, question generation, planning prompts or revision support. Learners should keep responsibility for checking claims and completing follow-up thinking.
2. Amber: AI-visible practice. AI may be used only at defined stages, such as before drafting or after an initial attempt. Learners record what help they used.
3. Red: independent evidence. No AI assistance while completing the task unless the task instructions explicitly permit it. This includes closed-book mock conditions where you want an exam-readiness measure.

For each resource or lesson, record four decisions: the learning objective; the activity label; permitted and prohibited AI use; and how the learner will demonstrate independent understanding. This record need not be bureaucratic. A line in a lesson plan or assignment brief is usually enough for internal practice.

Course creators should apply the same distinction to published resources. If a worksheet contains AI-generated questions, review them for factual accuracy, suitability and alignment with the intended specification before publication. If a model answer was developed with AI assistance, a subject expert should remain responsible for the final version.

How to design AI-resilient exam practice and feedback

AI-resilient practice does not depend on trying to outsmart every tool. It focuses on evidence that is hard to fake because it shows process, adaptation and live understanding.

• Use short, timed, no-AI tasks regularly rather than relying on one large mock.
• Ask learners to explain why they selected an answer, method or quotation.
• Set parallel questions after a supported example to test transfer.
• Collect planning notes, first attempts and revisions when the development process matters.
• Use brief oral check-ins: “Talk me through your first step” or “Why is this evidence relevant?”
• Give feedback on one or two high-value improvements, then ask for an independent rewrite or reattempt.

Feedback activities benefit from a clear division of roles. AI can suggest questions or flag possible gaps; the teacher decides what is educationally sound and what feedback to give. Learners should also understand that AI output may be incomplete or wrong, so it is not a substitute for subject knowledge, course materials or professional judgement.

Communicating expectations to learners and parents

Integrity expectations work best when they are taught, not merely announced. Explain that the purpose of independent practice is not to catch learners out. It is to identify what they can already do without help, so that tuition can focus on the right next step.

Use consistent language in lesson slides, course areas and assignment instructions. For example: “This is a red activity. Complete it without AI or outside help. We are using it to decide what to practise next.” For an amber activity, state the permitted point of use: “Draft your plan independently. You may then use AI to generate questions about missing counterarguments, and note any prompt you used.”

Parents may also need reassurance that responsible AI use is neither a shortcut to results nor an automatic problem. Explain the boundaries, the purpose of unaided practice and the fact that external assessment rules may differ by qualification and must be checked separately.

Downloadable checklist: AI assessment-integrity toolkit

Turn the following into a one-page downloadable handout for tutors, learners and course teams.

Lesson-planning decision tree

1. Is this activity intended to teach, practise or measure independent performance?
2. If it measures independent performance, is AI prohibited or expressly permitted by the relevant instructions?
3. If it teaches or practises, what specific AI role is allowed: explanation, questioning, planning, feedback or something else?
4. What must the learner do independently after using AI?
5. How will you check accuracy, authorship and understanding?
6. How will the activity be labelled: green, amber or red?

Student AI-use declaration template

Task: ____________________

Activity label: Green / Amber / Red

Did I use AI for this task? Yes / No

If yes, I used it for: explanation / questions / planning / feedback / other: ____________________

I did not use AI to complete any part marked as independent. Yes / No

What I changed or checked myself: ____________________

Learner name and date: ____________________

Pre-publication and pre-assessment checklist

• Is the purpose of the task explicit: learning, practice or independent evidence?
• Have you checked the current requirements from the relevant awarding organisation or centre where applicable?
• Is permitted AI use stated in the learner-facing instructions?
• Does the task include a way to see independent understanding?
• Have AI-generated materials been reviewed for accuracy, relevance and appropriateness?
• Would you interpret the result differently if AI had been used? If yes, are the conditions recorded?
• Do learners know how to ask for clarification before beginning?

SubSchool can help course teams turn repeatable routines such as activity labels, declaration prompts, feedback structures and revision checklists into consistent teaching workflows—while teachers retain authorship and the final educational decision. Explore SubSchool to see how you can reduce repetitive preparation without handing over assessment judgement.