Describing the output you want takes a paragraph. Showing it takes a single worked case. One Good Example is the idea behind few-shot prompting: include a worked input and its ideal output, and the model infers the format, tone and level of detail without a long rulebook.

Pick examples that represent the real cases, including an awkward one, and vary them so the model does not copy surface details. A few well-chosen examples often beat many. Re-test after any change, because the model follows your examples more faithfully than your adjectives.