Basic idea
A prompt works better when it gives the model a clear task, a clear audience, and a clear expected output. If the first answer is weak, it is usually better to refine the prompt than to start over with a completely different question.
Practical tips
- State the task clearly: explain what you want the model to do.
- Give the level and audience: for example, first-year university students or secondary school teachers.
- Specify the format: ask for bullet points, short explanations, exercises, or a step-by-step solution.
- Add constraints: mention length, notation, prerequisites, or topics to avoid.
- Ask for examples: request one or two concrete examples or counterexamples when helpful.
- Ask for checks: for mathematics, ask the model to highlight assumptions and possible uncertainties.
- Revise iteratively: if the answer is too advanced, too vague, or too long, say exactly what should change.
A simple pattern
I teach [course or level]. Create [type of material] on [topic] for [audience]. Keep the explanation [short/detailed], use standard notation, and include [examples/exercises/solution sketch]. Avoid [undesired feature].
Example: weak prompt vs better prompt
A generic prompt often produces a generic answer. A more specific prompt usually leads to something more useful.
Weak prompt: Explain derivatives.
Better prompt: Explain the derivative to first-year university students, using one geometric interpretation and two worked examples, without assuming prior knowledge of limits. Keep the explanation under 250 words and end with one short practice question.
Important warning
Even a well-written prompt does not make the output reliable. In mathematics and teaching, all claims, notation, and level choices still need to be checked by a human.
Further reading
For a beginner-friendly introduction with many examples, see Learn Prompting (external).