Get AI to interview you before it writes anything

Get AI to interview you before it writes anything

Most AI output for Australian teachers reads like a US worksheet because the teacher gave a half-context prompt and the model filled in the missing bits with the average of everything it has ever read. The missing bits are almost always the Australian ones. Year level, ability spread, ACARA reference, the two boys in the back row, the fact that you have thirty-eight minutes on a Thursday and not sixty on a Monday. First Saturday of the winter holidays is a good time to internalise the technique that fixes this once, not per-prompt. You flip the model into interview mode and let it ask you before it writes.

The prompt

You are going to help me with a classroom task, but I want you to interview me first.

Here is what I want:
- Task: [e.g. plan a 60-minute lesson, write a rubric, draft a family email, sequence six lessons, design an extension task]
- Year level and subject: [e.g. Year 9 Australian History]
- School context: [e.g. metropolitan Victorian state secondary, mixed ability, one EAL student and two on individual learning plans]

Before you produce anything, ask me between three and five questions that will materially change the output. Prioritise questions about:
1. What my students already know, and the specific misconceptions or errors they typically bring to this topic
2. Constraints I have not mentioned (time on task, resources, wellbeing, EAL and LD needs, behaviour patterns, timetable quirks)
3. What "good" looks like from my point of view, not the textbook's

Ask the questions as a short numbered list. Do not produce the task output until I have answered all of them. If any of my answers is thin or generic, push back once with a follow-up question before moving on.

When you do produce the output, use Australian English, align to ACARA v9.0 (or the relevant VCE, VCE-VM, HSC or IB syllabus) where relevant, and cut anything that reads like a US worksheet.

What it does well

Forces the model to gather the context it usually invents. Single-shot prompts produce generic output because the model has to make assumptions and it makes American ones by default. Interview mode is the difference between a lesson plan you rewrite from scratch and one you tweak for ten minutes and use.

Where it falls over

Smaller and cheaper models will sometimes still shortcut past the interview and start producing. If that happens, reply with "Stop. Ask me the questions first" and it usually corrects. The other failure is on you: the interview only helps if you answer the questions properly. Typing "yes" and "same as usual" gets you the same generic output as before, just after a longer wait.

Try it like this

For a Term 3 unit: set Task to "sequence six lessons on [topic]" and let it interview you before it drafts week one. For a mid-year report comment bank: set Task to "draft four comment variations for a mid-band Year 8 English student, differing in tone and specificity". For a summative task you have been putting off: set Task to "design a summative assessment task for [subject] on [topic] with a rubric that discriminates between B and A". If you only have five minutes: skip the full interview and add "before you write, list the three assumptions you are making that would most change the output" instead.

Save this one somewhere you will find it. It is the move that makes every other prompt in this newsletter work harder.