How do I write better prompts to get useful answers from AI?

Generic prompts get generic answers because you've given the model nothing to narrow toward. The fix is almost always more context and a clearer specification of what 'good' looks like — not clever phrasing tricks. The five changes that produce the biggest improvement: 1. **Give context about your situation.** 'How do I improve my resume' has no anchor. 'I'm a self-taught developer with no professional experience applying for junior React roles in India, here's my resume, what's weakest?' gets a specific answer. The model can only tailor to what you tell it. 2. **State the goal, not just the task.** 'Write a cover letter' versus 'Write a cover letter that emphasises my open-source work, since the job posting stresses collaboration'. The second contains the criterion for success. 3. **Specify format, length and audience.** 'Explain caching' will give you an essay. 'Explain caching in 150 words to a junior developer who knows basic web development, with one concrete example' gives you something usable. Vague requests get hedged, average-length, average-depth output because that's the safest prediction. 4. **Provide examples of what you want.** Pasting two examples of the style you're after does more than three paragraphs describing it. This is the single most underused technique. 5. **Iterate instead of restarting.** Treat the first output as a draft and push back: 'too generic, use my actual project details', 'cut the introduction', 'you missed the security implications'. The best results almost always come from three exchanges, not one perfect prompt. Techniques worth knowing: - **Assign a role** when it genuinely constrains the answer: 'act as a code reviewer focused on security' meaningfully changes the output. Generic role-play ('you are a world-class expert') mostly doesn't. - **Ask for reasoning before the conclusion** on analytical problems — working through steps before answering measurably improves accuracy. - **Ask for multiple options** — 'give me three approaches with trade-offs' beats one confident recommendation when you're deciding something. - **Ask it what it needs.** 'What information would help you answer this better?' is a genuinely good move when you're unsure what to include. What doesn't work: threats, fake urgency, offering tips, elaborate 'you must' framing. And the important discipline — verify anything factual. A confidently wrong answer looks identical to a correct one, so the better your prompts get, the more careful you should be about believing the fluent output.