You don't need a hundred data work prompts. You need five you trust and actually reuse.
Reuse the recipe
Lock the parts that never change and leave one slot for the variable. Now every run shares a family resemblance instead of starting from scratch.
Keep a negative list
Write down what you never want as explicitly as what you do. The exclusions often do more work than the instructions.
- Compare two outputs instead of trusting one
- Have the model flag what it could not cover
- Keep one ask per prompt — chaining beats cramming
Cut before you polish
Ask for the reasoning, then the result. When the model shows its steps you can see exactly where to push back.
Ask for the reasoning, then the result. When the model shows its steps you can see exactly where to push back.