The Playbook
Download the playbook (PDF) ↓For teaching tasks such as creating an assignment, improving a worksheet, finding subject connections or designing feedback.
Structure: How to write a useful AI specification
Before asking AI to create something, give it enough direction to do the job well. A specification is simply a clear brief. These five questions help you write one.
Prompt vs. specification: A prompt tells AI what to produce; a specification defines how the work should be done.
| Question | What to tell AI | Example: an assignment for a project on freedom |
|---|---|---|
| 1. Task: What do you want AI to do? | Explain the job, why it matters and what you want back. | "Suggest two English assignments for Year 7 pupils. Each should help them use evidence to explain a point about freedom." |
| 2. Materials: What should AI work with? | Provide relevant curriculum goals, existing work, pupil context and research. Check that AI can actually read them. | "Use the attached English standards, project brief and reading extract. Tell me if you cannot access any of them." |
| 3. Process: How should AI approach the task? | Give important steps and limits. Say what to do when something is missing. | "Read the standards first. Link each idea to a pupil action and visible evidence. Don't invent curriculum requirements." |
| 4. Quality: What would make the result good? | State criteria you can check. Add a good and poor example, with explanations, if you have them. | "Pupils must make a claim and support it with evidence. A poster that only repeats facts is not enough." |
| 5. Approval: Who decides? | Say what AI may suggest and what a teacher or team must approve. | "AI proposes. I check the sources and fit for my pupils. The subject team approves curriculum links before use." |
These are five parts of a brief, not a sequence of steps. Write them in any order and leave out what adds nothing. For important teaching decisions, make Quality and Approval explicit.
Process: How to build a better specification through research and design
The four stages fit into two jobs: Research and Design. Each has an open-up step and a narrow-down step. You can go back when new information changes your mind.
Following instructions vs. good teaching: AI can follow a specification perfectly and still produce pedagogically weak results. That's why we need a better process.
| Research (Discover + Define) | Design (Develop + Deliver) | |
|---|---|---|
| Goal | Understand the teaching problem and decide what a good solution needs to achieve. | Create something that meets the agreed brief, then test and improve it. |
| Open up | Discover: Look at your pupils, curriculum, existing materials and relevant research. Gather best-practices and ask what you might be missing. | Develop: Write out a more detailed specification based on your findings, instructions and source material. Ask for different ideas or one small example. |
| Narrow down | Define: Choose the insights that matter, curate your findings. E.g. decide what pupils should learn, what they should show and what must stay unchanged. | Deliver: Compare the result with your quality criteria. Test edge-cases of your workflow. Check sources and classroom fit. Decide use, fix or drop. |
| What you leave with | A clear, evidence-informed specification input for the design phase. | A repeatable workflow that produces validated results meeting your quality standards. |
| What this avoids | Letting AI create materials before defining the right problem, allowing it to make key decisions for you. | Accepting an attractive AI output without knowing whether it works for your pupils. |
Research does not always mean finding academic papers. Reading the actual curriculum, an existing worksheet or pupils' previous work may tell you what needs changing. Use research tools when outside evidence could improve a decision; check the original sources.
Best-practices: Six practical techniques to improve your specification
Pick a technique when you need it. These are follow-up requests you can adapt to almost any teacher task.
| Technique | Copy-ready prompt | Helps when… |
|---|---|---|
| 1. Challenge my assumptions | "I think the problem is [X]. What are two other plausible explanations? What evidence would help us choose?" | You may have framed the wrong problem. |
| 2. Ask better questions | "Before you proceed, ask me up to three questions that could substantially change the result. Ask them one at a time." | AI would otherwise fill gaps with guesses. |
| 3. Check the evidence | "For each important claim, cite the supporting source. Explain how the evidence supports your decisions, and flag anything you cannot verify." | A convincing answer may be unsupported. |
| 4. Make quality checkable | "Turn words such as 'engaging', 'authentic' and 'interdisciplinary' into criteria I can inspect. What must pupils actually do or show?" | Your expectations are too vague to guide or assess the work. |
| 5. Preview before creating | "Based on my brief, tell me exactly what you would make, what pupils would do and what you would have to assume. Stop and wait for my approval." | You want to catch misunderstandings before generating a full resource. |
| 6. Test a small version and find the problem | "Make one example, not the full pack. Check it against my criteria. If it falls short, is the problem in the brief, sources, instructions or proposed activity? Suggest one change and a way to test it." | Iterating has become endless rewriting. |
Examples help: If you have a strong example and a weak one, show both and explain the difference. If you don't, use clear criteria rather than pretending an example is school-approved.
When the output disappoints: Use it if it meets the brief and your checks. Fix the specification or the output when you can identify a repairable problem. Drop it if the proposed approach does not serve the learning goal. Retest the changed part.
For your practice task: choose a teaching problem that matters to you. Bring back your brief, a small AI-generated example and one decision you can explain. You do not need to finish a complete teaching pack.
Privacy: use fictional pupil examples or properly anonymised materials in permitted systems. Never upload identifiable pupil work to an unapproved AI tool.
Your completion ticks stay in this browser.