How to Write AI Prompts That Get Better Answers

There is no secret phrase that unlocks a chatbot. The people who get the best results do something ordinary: they explain what they want the way they would to a capable colleague who knows nothing about the situation. This guide gives you a method that works with ChatGPT, Claude, Gemini and any other assistant.

The five parts of a good prompt

1. Goal — what you want done. Use a clear verb: summarise, compare, draft, rewrite, explain, critique.

2. Context — what the model needs to know. Background, constraints, what you have already tried, the material to work from.

3. Audience — who it is for. A board, a customer, a ten-year-old, yourself.

4. Format — what the answer should look like. Length, structure, tone, a table or bullet points, what to leave out.

5. Example — what good looks like. One sample of the style you want is worth a paragraph of description.

You do not need all five every time. Add more when the answer matters more.

Before and after

Weak: "Write an email about the delay."

Better:

Draft an email to a client, Maria, explaining that her website launch will be one week late, moving from 3 March to 10 March. The cause is a delay in receiving her product photos. Tone: professional and warm, taking responsibility without blaming her. Offer a short call this week. Under 120 words. Give me two versions: one more formal, one more relaxed.

The second prompt states the goal, the facts, the audience, the tone, the length and the output. The reply will need little editing.

Weak: "Explain inflation."

Better:

Explain inflation to a 16-year-old who has just started their first part-time job. Use one everyday example involving wages and prices. Keep it under 200 words and avoid jargon; if you must use a technical term, define it.

Weak: "Summarise this report."

Better:

Summarise the attached report for a manager who has five minutes. Give me: the three main findings, the one decision it asks us to make, and any risks it mentions. Use bullet points. Quote the page number for each finding. If something is unclear in the report, say so instead of guessing.

Techniques that reliably help

• Give it the source. Paste the text or upload the file and say "use only this". It is the best defence against invented facts; see AI hallucinations explained.
• Ask for steps on hard problems. "Work through this step by step before giving your answer."
• Ask for options. "Give me three approaches with the trade-offs of each."
• Assign a perspective. "Review this as a sceptical investor would." It changes what the model pays attention to.
• Set limits. Word counts, reading level, what to exclude.
• Ask it to ask you. "Before you start, ask me any questions you need answered." This is especially useful for big tasks.
• Say what you do not want. "No introduction, no summary at the end."
• Invite doubt. "Tell me which parts you are least sure about."

Iterate

The first reply is a draft. The conversation is where the quality comes from.

• "Shorter, and cut the second paragraph."
• "More direct. Remove the hedging."
• "Keep the structure but rewrite it for a non-technical reader."
• "That is wrong about X — here is the correct detail. Redo that section only."

Be specific about what to change and what to keep. If a conversation has drifted, start a new chat with a short summary; long threads lose track of early instructions, for reasons explained in tokens and context windows.

A reusable template

Task: [what you want] Context: [background, constraints, source material] Audience: [who will read it] Format: [length, structure, tone] Example: [a sample, if you have one] Checks: Ask me questions first if anything is unclear. Flag anything you are unsure about.

Save it as a note and fill it in for anything important.

Common mistakes

• Too little context. The model cannot read your mind or see your screen.
• Several tasks in one message. Split them.
• Vague quality words. "Make it better" gives the model nothing to aim at. Say better how.
• Accepting the first answer.
• Trusting specifics. Names, numbers, quotations and references need checking.
• Pasting confidential material without thinking. See what not to share with AI chatbots.

Prompts for agents

When the AI will act instead of just answer, add two things: limits ("spend no more than £50") and stopping points ("show me the options and wait for approval"). More in what are AI agents.

Does the model matter?

Yes, but less than the prompt. A clear prompt on a free model usually beats a vague one on the most expensive. If you are choosing between assistants, see ChatGPT vs Claude vs Gemini; for pictures, the rules differ and are covered in how to write prompts for AI image generators.

Frequently asked questions

What makes a good AI prompt?

A clear goal, enough context, a defined audience, a stated format and, where possible, an example of what you want.

Do I need to say please to a chatbot?

Politeness does not hurt, but clarity is what improves the answer.

How long should a prompt be?

As long as it needs to be. A few sentences is fine for simple tasks; a detailed brief is worth writing for important ones.

Why does the AI ignore part of my prompt?

Usually because the request was buried in a long message or conflicts with another instruction. Put key instructions at the end and split large tasks.