Methods · Working with the tool itself · Short guide

Prompt iteration as thinking

Treat refining the question as the intellectual work, so each rewrite sharpens what you actually need to know, not just how you boss the model.

When to use this: While you practise
Learning impact: Clear help, modest gain
Learning evidence: Thin or mixed evidence
Where you are with the topic: Building understanding
How often to use it: Worth repeating often

When to use it

Your first prompt got a vague essay and you are about to regenerate blindly.
You cannot yet say in one sentence what would count as a good answer.
You are iterating wording for polish instead of clarifying the goal.

What to ask AI

Type something like: “Here is my draft prompt: [ ]. Do not answer the subject yet. Interview me to improve the prompt.”

Ask it to press on goal, constraints, what you tried, and what success looks like.

Keep the hard part on your side: you edit the prompt. The model is a pressure surface, not the thinker.

How it helps

Often the intellectual work is clarifying the question, not consuming the first answer.

Each revision can sharpen what you actually need to know.

If you only reword for tone, you are polishing, not thinking.

Treat prompt iteration as problem formulation: what would count as success?

How to run it

Paste your draft prompt and forbid subject answers until the prompt is sharp.
Answer its clarifying questions honestly: goal, level, constraints, failed attempts.
Rewrite the prompt yourself from that interview.
Keep a short diff log: what each revision was trying to fix.
Only then run the improved prompt. Compare output quality to the first try.

Example exchange

You

Here’s my draft prompt: “Explain photosynthesis.” Don’t answer the subject yet. Help me improve the prompt first.

What do you need to be able to do afterward, exam explain, lab mechanism, or teach a friend?

You

Exam explain: why light reactions precede the Calvin cycle, in our course vocabulary, under eight sentences, no children’s analogy.

Then your prompt should name that success check, the vocabulary constraint, and the forbidden analogy. Type the revised prompt yourself, then we can run it.

Copyable prompt

Here is my draft prompt: [ ].
Don’t answer the subject yet. Interview me to improve it, goal, constraints,
what I’ve tried, and what would count as success, one question at a time.
Then show a revised prompt for me to edit.

The Tell

Here is how you know this method has flipped: ten prompt versions later, you still cannot state the question in one clear sentence yourself.

Then you were cycling cosmetics, not formulating the problem.

Stop regenerating. Write the success check first, then one prompt that serves it.

Principle evidence

Strength of the underlying learning idea, not a claim about AI products.

The underlying learning idea is rated weak-to-moderate. Problem formulation and clarifying what would count as an answer support deeper learning more than cosmetic rewording. Treating prompt revision as the thinking surface is a plausible transfer of that idea. Prompt-engineering-as-pedagogy remains mostly speculative.

AI delivery evidence

Whether an AI tutor delivers this method well is a separate question.

No claim here that AI interviews reliably teach formulation skill. Delivery is speculative. Your rewritten success criteria are the learning product.