Why does the same AI give others great results but not you? It's the prompt. This gives a reusable framework, six advanced techniques, common mistakes and office rewrite examples.
Five parts of a good prompt
A clear prompt fills in the context AI lacks. Remember the skeleton: role + task + context + output format + constraints. The more complete these are, the less rework you'll do.
Six techniques that work instantly
- Give examples (few-shot): one or two input→output samples beat paragraphs of description;
- Ask it to reason: for hard problems add "think step by step, then conclude";
- Set role & audience: "explain to a total beginner" is more controllable than "explain";
- Specify structure: request headings, tables or numbering for easy reuse;
- Iterate: don't restart — say "keep X, make Y more concise";
- Add a negative list: "no fluff, no chatty endings, don't invent data".
An office rewrite example
Before: "write a notice".
After: "You're an HR officer. Draft a National Day holiday notice for all staff, covering dates, on-call duty and reminders, formal and concise, under 200 words, key info as a bullet list." Locking role, audience, structure, length and format gets you there in one pass.
Four common mistakes
- Too vague: missing context and format yields generic output;
- Too many goals at once: split into turns;
- No verification: AI fabricates confidently — check facts and numbers;
- No reuse: save good prompts as templates.
