Strong prompts combine task, audience, context, constraints, format and a check.
Do now: Label those six parts in one prompt you already use.
Include a component only when it changes quality or safety.A repeatable structure lets you improve prompts systematically instead of collecting mysterious phrases from the internet.
The six working parts
Task defines the action. Audience defines who will use the answer. Context supplies relevant facts. Constraints set boundaries such as time, length, data or tone. Format shapes the return. Check asks for evidence, uncertainty or self-inspection.
Order is flexible; clarity is not. Put the most important instruction where it is easy to find and separate source material with clear headings or delimiters.
Structure helps both the model and the human reviewer.
Criteria beat decoration
Replace ‘professional’ with observable criteria: direct subject line, under 140 words, one clear request and no unsupported claims. Replace ‘make it African’ with the actual location, audience, language and lived context required.
Negative constraints are useful when specific: do not invent figures; do not change quoted text; do not include personal data.
Diagnose the miss
If the answer solves the wrong problem, repair the task. If it is generic, add relevant context. If it is hard to use, specify format. If it overreaches, add boundaries and checks.
Change one or two parts at a time so you learn which instruction improved the result.
Prompt debugging is easier when the parts are visible.