Specify the finish line
Define the audience, scope, length, tone, and quality bar. Say what a successful output must include and exclude.
The library
Start with a tested structure, then shape it around your data, constraints, and definition of a good result.
Prompt engineering basics
Good prompts remove guesswork. These seven habits make outputs easier to evaluate, reproduce, and improve.
Show the format
Give a compact example when structure matters. A real input and ideal output can resolve ambiguity faster than another paragraph of instructions.
Separate content with delimiters
Wrap source material in clear tags or fences. This helps distinguish instructions from the text being processed.
Assign a useful role
Use a role to set expertise and perspective, not for decoration. Name the decisions that role is expected to make.
Request structured output
Use a schema, keys, allowed labels, or a fixed outline when another person or system needs to consume the result reliably.
Test edge cases
Try missing fields, conflicting facts, long inputs, and ambiguous examples. Revise the prompt where results become inconsistent.
Version every change
Keep the prompt, test inputs, expected behavior, and revision notes together. Compare versions against the same evaluation set.