How to Write Better AI Prompts: A Practical Guide With Examples

Vague prompts get generic answers. Learn the six building blocks of a good AI prompt, with before-and-after examples and a copy-paste template that works in ChatGPT, Claude and Gemini.

Overhead view of a woman typing on a silver laptop at a small table with a cup of coffee and flowers
Overhead view of a woman typing on a silver laptop at a small table with a cup of coffee and flowers
In this article
  1. Why most AI prompts disappoint
  2. The 6 building blocks of a good prompt
  3. How to write better AI prompts: step by step
  4. 1. Start with the task and the goal
  5. 2. Add the context only you know
  6. 3. Specify the format
  7. 4. Show, don't just tell
  8. 5. Give it a role when it helps
  9. 6. Break big jobs into steps
  10. 7. Let the AI ask you questions
  11. 8. Iterate instead of starting over
  12. 9. Ask it to check its work
  13. A reusable prompt template
  14. Three more before-and-after examples
  15. Mistakes to avoid
  16. Frequently asked questions
  17. Do the same prompts work in ChatGPT, Claude and Gemini?
  18. Are longer prompts always better?
  19. Does being polite to AI change the answer?
  20. Do I need to learn "prompt engineering" to use AI well?
  21. Why does the AI give a different answer each time?
  22. The bottom line

The secret to writing better AI prompts is simple: treat the chatbot like a smart new colleague who knows nothing about your situation. Tell it what you want, why you want it, who it's for and what the result should look like, then refine the answer through follow-ups instead of starting over. The steps below work in ChatGPT, Claude, Gemini, Copilot and most other assistants, and include before-and-after examples you can copy.

Why most AI prompts disappoint

A vague request gets a generic answer. When you type "write a cover letter" or "explain inflation", the AI has to guess your audience, tone, length and purpose, so it falls back on the most average version possible. Nearly every improvement below comes down to removing guesswork.

The 6 building blocks of a good prompt

ElementWhat to includeExample
TaskThe specific action, starting with a clear verb"Summarise", "Draft", "Compare", "Rewrite"
ContextBackground the AI can't know"I run a two-person bakery and we're raising prices 10%."
Audience and goalWho will read it and what should happen next"For regular customers. I want them to stay loyal."
FormatLength, structure, style"Under 120 words, friendly, no exclamation marks."
Examples or source materialText to work from or a sample of the style you want"Here's last year's newsletter for tone: ..."
ConstraintsWhat to avoid or must include"Don't mention competitors. Include our opening hours."

You don't need all six every time. A quick factual question needs none of them. A piece of writing you'll send to other people benefits from most.

How to write better AI prompts: step by step

1. Start with the task and the goal

Lead with what you want and why. "Help me write an email" is a start. "Write an email asking my landlord to fix a broken boiler, firmly but politely, so it gets fixed this week" gives the AI a goal to aim for.

2. Add the context only you know

The AI doesn't know your job, your audience, what you've already tried or what "good" looks like to you. A couple of sentences of background often improves the answer more than anything else.

Before: Give me a workout plan.

After: I'm 42, haven't exercised in two years and have a bad left knee. I can train three times a week for 30 minutes at home with a pair of dumbbells. Give me a four-week beginner plan that avoids lunges and jumping, as a simple table.

3. Specify the format

Ask for exactly the shape you need: a bulleted list, a table, three options, a 150-word paragraph, an email with a subject line, or a step-by-step checklist. If you plan to paste the result somewhere, say where ("as a LinkedIn post", "as plain text for a text message").

4. Show, don't just tell

If you want a particular style, paste an example. One or two samples of the tone, structure or level of detail you like are often more effective than describing it. The same goes for source material. If you want a summary of a report, paste the report (or upload it, if the tool supports files) rather than asking the AI to recall it.

5. Give it a role when it helps

Phrases like "You are an experienced editor" or "Act as a patient maths tutor for a 12-year-old" can steer vocabulary, depth and tone. Roles work best when combined with the context above, not as a replacement for it.

6. Break big jobs into steps

Instead of "Write my business plan", work in stages: brainstorm sections, then outline one section, then draft it, then critique it. Each answer becomes context for the next request, and you can correct course early.

7. Let the AI ask you questions

When you're not sure what to include, end your prompt with: "Before you start, ask me any questions you need answered to do this well." This turns a one-shot guess into a short interview, and the result is usually far more tailored.

8. Iterate instead of starting over

Treat the first answer as a draft. Follow-ups such as "Make it shorter", "The second point is wrong because…", "Use simpler language" or "Give me three alternative headlines" are faster than rewriting your prompt from scratch. If a conversation drifts badly off course, though, a fresh chat with a better prompt can work better.

9. Ask it to check its work

Requests like "List any assumptions you made", "Point out the weakest part of this argument" or "Double-check these calculations step by step" often surface problems. They don't guarantee accuracy, but they make errors easier to spot.

A reusable prompt template

Copy this, fill in the brackets and delete what you don't need:

Task: [what you want, starting with a verb]
Context: [background, what you've tried, relevant facts]
Audience: [who it's for and what they should do or feel]
Format: [length, structure, tone]
Examples / source: [paste text or a style sample]
Constraints: [must include / must avoid]
Before you begin, ask me any clarifying questions.

Three more before-and-after examples

Before: Summarise this article.

After: Summarise the article below for a busy manager in five bullet points. Put the single most important takeaway first, and flag any claims that aren't backed by data in the article.

Before: Plan a trip to Lisbon.

After: Plan a relaxed three-day Lisbon itinerary for two adults in their 60s who dislike steep walks. We like food markets, tiled architecture and one museum a day at most. Include a rough walking or tram plan for each day.

Before: Fix my spreadsheet formula.

After: In Google Sheets, column B has dates and column C has sales. I want the total sales for the current month in cell F1. My formula =SUMIF(B:B, MONTH(TODAY()), C:C) returns 0. Explain why and give me a corrected formula.

Mistakes to avoid

  • Trusting facts blindly. AI models can state wrong information confidently, including invented sources, quotes and statistics. Verify anything important, especially medical, legal, financial or news-related claims.
  • Sharing sensitive data. Don't paste passwords, ID numbers, or confidential work or client information unless you understand how your tool and your employer handle that data. Check the privacy and training settings in your AI app.
  • Overstuffing one prompt. Ten unrelated requests in one message produce shallow answers to all of them.
  • Using negative-only instructions. "Don't be boring" is less useful than "Use short sentences and one concrete example per point."

Frequently asked questions

Do the same prompts work in ChatGPT, Claude and Gemini?

Mostly, yes. The principles of clear tasks, context, format and examples apply to all major assistants. Each vendor publishes its own guidance, such as Anthropic's prompt engineering overview, OpenAI's prompting guide and Google's Gemini prompting strategies, and they overlap heavily.

Are longer prompts always better?

No. Longer prompts help when the extra words add useful context or constraints. Padding, repetition and contradictory instructions make results worse.

Does being polite to AI change the answer?

Politeness doesn't hurt, but clarity matters far more. A clear, specific request will beat a vague but courteous one.

Do I need to learn "prompt engineering" to use AI well?

Not for everyday tasks. The habits in this guide cover most needs. Advanced techniques matter more for developers building apps and AI agents that run prompts automatically.

Why does the AI give a different answer each time?

Most chatbots include some randomness so responses don't sound robotic. If you need consistency, give more specific instructions and examples, or ask for a fixed format.

The bottom line

Better AI prompts come down to removing guesswork: say what you want and why, add context, specify the format, show an example, then refine through follow-ups. Always check important facts yourself. To put these skills to work, see our roundup of the best AI tools for productivity, and explore more in our AI section.

Add The Tech Wonder as a preferred source on Google

The Tech Wonder Team

Editorial team

The Tech Wonder covers technology news and writes practical, plain-English guides to the gadgets, apps and ideas shaping everyday life.

0 Comments

Join the discussion. Comments are moderated and your email is never published.