Most effective prompts share the same building blocks. Missing one is the usual cause of disappointing output.
The Building Blocks
- Role and context: who the model is helping and why.
- Task: the specific output wanted.
- Inputs: the material to work on, clearly delimited.
- Constraints: length, tone, scope, what to avoid.
- Output format: structure, headings, JSON, a template.
- Quality bar: what a great answer looks like, or examples.
- Uncertainty handling: what to do when information is missing.
A Reusable Template
Context: I'm <role> working on <situation>. The output is for <audience>.
Task: <exactly what to produce>.
Material:
<document>
...
</document>
Constraints: <length, tone, must-include, must-avoid>.
Format: <structure>.
If something is unclear or missing, <what to do>.
Specific Beats General
"Write a 150-word summary for customers who aren't technical, highlighting what changes for them" beats "summarise this".
Explain Why
Reasons behind constraints help the model make good choices in cases your rules don't cover.
Iterate
Treat prompts as drafts. Look at outputs, identify what's wrong, and adjust the specific building block responsible.