How-To

How to Write Better AI Prompts: A Practical Guide With Before-and-After Examples

Six habits that get noticeably better answers from ChatGPT, Claude, Gemini and other assistants, with real before-and-after prompts you can copy.

In this guide
  1. 1. Say who it's for and why
  2. 2. Describe the result you want
  3. 3. Give it the material
  4. 4. Show an example
  5. 5. Break big jobs into steps
  6. 6. Treat the first answer as a draft
  7. Always check the facts
  8. A reusable prompt template
  9. Does the model matter?

Two people can use the same AI model and get completely different results. The difference is rarely the model. It's usually the request. The good news is you don't need tricks or "magic words." You need to give the AI the same information you'd give a smart new coworker on their first day.

Here are six habits that make the biggest difference, with examples.

1. Say who it's for and why

AI doesn't know your situation unless you tell it. The single biggest improvement most people can make is adding context: who the result is for, what it will be used for, and what you already know.

Before:

Write an email about the office closing early.

After:

Write a short, friendly email to a 12-person office telling everyone we're closing at 2 p.m. this Friday for building maintenance. Remind them to save their work and take laptops home if they plan to work from home. Keep it under 100 words.

The second prompt takes 20 more seconds to write and usually produces something you can send without edits.

2. Describe the result you want

Tell it the format, length and tone. If you want a table, a bulleted list, three options or a one-paragraph summary, say so.

Instead ofTry
"Summarize this.""Summarize this in five bullets for a busy manager. Put any deadlines first."
"Help me plan a trip.""Make a day-by-day table for 4 days in Boston in May: one main activity, one place to eat, and travel time between them."
"Make it better.""Make this more concise and confident. Keep my main points and stay under 150 words."

3. Give it the material

Assistants are much more accurate when they work from text you provide rather than from memory. Paste the document, the email thread or the notes (after removing anything sensitive; see what not to share with AI chatbots) and ask it to use only that material.

Using only the meeting notes below, list every action item, who owns it, and the due date. If a due date isn't mentioned, write "not set."

That last instruction matters. Telling the AI what to do when information is missing reduces the chance it makes something up.

4. Show an example

If you have a specific style in mind, one example beats a paragraph of description.

Write five more product descriptions in the same style as this one: "Hand-poured soy candle with notes of cedar and orange. Burns for about 40 hours. Made in small batches in Vermont."

The AI will match the length, structure and tone far more reliably than if you'd written "short, warm, factual descriptions."

5. Break big jobs into steps

Asking for a complete business plan, a full research report or a whole website in one message usually gets you something generic. Work in stages instead:

  1. "Ask me 10 questions about my business so you can help me write a one-page plan."
  2. Answer the questions.
  3. "Now draft an outline."
  4. "Expand section 2, and keep it under 300 words."

Each step lets you correct course before the next one, and the final result is much more specific to you.

6. Treat the first answer as a draft

The first response is a starting point. Reply with specific feedback, just as you would with a person:

  • "Good, but shorter. Cut it in half."
  • "The second option is closest. Make it less formal."
  • "You missed the deadline in the third paragraph of my notes. Add it."
  • "What are the weakest points in this argument?"

Asking the AI to critique its own work, or to argue the other side, is a quick way to find gaps.

Always check the facts

Even the best models sometimes state wrong information confidently. This is often called "hallucination." Before relying on an answer for anything that matters:

  • Ask for sources, then open them. Some links may not exist or may not say what the AI claims.
  • Double-check numbers, dates, names, quotes and laws. These are where errors show up most.
  • Use assistants with web search for anything recent. A model's built-in knowledge stops at its training cutoff.
  • Be extra careful with health, legal and financial questions. Use AI to understand and prepare, then confirm with a professional.

A reusable prompt template

When you're not sure where to start, fill in this template:

Role: You're helping me as a [e.g. editor, tutor, planner]. Task: I need [the thing you want]. Context: It's for [who], who [what they need to know or do]. Here's what I have so far: [paste material]. Format: Give it to me as [bullets, table, email, three options], about [length]. Rules: Use only the material I gave you. If something is missing, tell me instead of guessing.

Does the model matter?

Yes, but less than you'd think for everyday tasks. A well-written prompt to a mid-priced model often beats a vague prompt to the most expensive one. For hard reasoning, long documents or coding, the top models do pull ahead. Check this week's AI standings to see which models lead each category, or take the Which AI should I use? quiz.

Keep reading