How to Write Good AI Prompts
ChatGPT, Claude and Gemini give much better answers when you ask well. Learn the handful of habits that make the difference, build a strong prompt with the builder, and check your own prompts against a simple checklist.
Why a good prompt matters
A prompt is whatever you type into an AI chatbot. The chatbot can't see your screen, doesn't know who you are, and can't read your mind. All it has is your words. If you ask a vague question, it fills the gaps with a generic guess.
Think of it like asking a smart new coworker for help on their first day. They're quick and know a lot, but you still have to explain the job.
“Write an email about the meeting.”
Which meeting? To whom? Formal or casual? You'll get a bland template you have to rewrite.
“Write a short, friendly email to my team…”
“…saying Thursday's 2 pm budget meeting moved to Friday at 10 am in Room B. Ask them to reply if they can't make it. Under 100 words.”
The good news: you don't need special tricks or “magic words.” The official guides from OpenAI, Anthropic and Google all say the same thing: be clear, be specific, and give the context a person would need.
Give it context and be specific
Most great prompts answer the same few questions. You don't need all of them every time, but the more the answer matters, the more you should include.
- Role or situation. Who are you, or what should it act as? “I'm a first-time home buyer.” “Act as a patient math tutor.”
- The task. One clear verb: write, explain, summarize, compare, plan, fix, list.
- Background. The facts it can't know: your budget, your city, the deadline, what you already tried.
- Audience. Who will read the result? “For my 10-year-old.” “For my boss, who hates jargon.”
- Limits. Length, tone, what to avoid. “Under 150 words.” “No bullet points.” “Vegetarian.”
- Weak: “Give me dinner ideas.”
- Better: “Give me 5 weeknight dinner ideas.”
- Strong: “I cook for a family of four, including a picky 7-year-old. Give me 5 weeknight dinners that take under 30 minutes, use one pan, and cost less than $15. No seafood.”
Paste the material in. If you want it to work on something, include it: the email you're replying to, the paragraph to shorten, the error message you see. Putting it between quote marks or under a line like “Text:” helps it tell your instructions apart from the material.
Show an example and ask for a format
Examples beat descriptions. If you want a certain style, show one. “Write it like this:” followed by a sample works better than a paragraph explaining the style. One or two examples are usually enough.
“Suggest 10 names for my dog-walking business. I like short, playful names like Pawsome Walks and Leash Go. Avoid puns about ‘ruff’.”
Ask for the shape you want. Chatbots can answer as a table, a checklist, numbered steps, a short paragraph, an email, or even a spreadsheet-ready list. Say so up front:
| If you want… | Add this to your prompt |
|---|---|
| Something to compare | “Put it in a table with columns for price, pros and cons.” |
| Instructions to follow | “Give me numbered steps, one action per step.” |
| A quick answer | “Answer in 2 sentences, then stop.” |
| Something to paste | “Only give me the email text, no introduction.” |
| Simple language | “Explain it like I've never heard of it. No jargon.” |
Tell it what to do, not only what not to do. “Write in plain, everyday words” works better than “don't be too technical,” because it gives the chatbot a clear target.
Build a prompt, then check one
Fill in what you know. Leave a box empty to skip it. The builder turns your answers into one clear prompt you can copy into any chatbot.
The checker looks for clues in your wording. It's a rough guide, not a grade: a short prompt can be perfect for a simple question.
Treat it like a conversation
The first answer is a draft. The real skill is the follow-up. The chatbot remembers what you said earlier in the same chat, so you can steer it without starting over.
- Say what to change. “Shorter.” “Make the second paragraph friendlier.” “Swap the chicken recipe for something vegetarian.”
- Say what was good. “Keep the opening line, redo the rest.” This stops it from throwing away the parts you liked.
- Let it ask you. For bigger jobs, end with “Ask me any questions you need before you start.” It will often ask about things you forgot to mention.
- Ask for options. “Give me 3 different versions” is a fast way to find a direction you like.
- Break big jobs into steps. Outline first, then each section. You'll catch problems early.
- Start a fresh chat for a new topic, or when it keeps repeating a mistake. Old instructions in a long chat can pull answers off track.
- “Write a thank-you note to my neighbor for watering my plants while I was away.”
- “Nice. Make it shorter and mention the tomatoes she saved.”
- “Perfect. Now a version I can text instead of writing on a card.”
Check the facts and protect your privacy
Chatbots can be confidently wrong
AI chatbots can make up facts, dates, quotes, links and even book titles. OpenAI calls this hallucination, and says a chatbot can sound certain while being wrong. Treat answers as a first draft, especially for anything important.
- Ask for sources, then open the links yourself. Make sure the page exists and actually says what the chatbot claims.
- Double-check numbers, dates, quotes and names. These are the most common slip-ups.
- Know its limits. Without a web search tool, a chatbot only knows what it learned up to its training cutoff, so recent news, prices and rules may be out of date.
- Health, legal, tax and money decisions: use AI to understand the topic and prepare questions, then confirm with a doctor, lawyer, tax pro or the official source.
Keep private things out of the chat
What you type is sent to the company's servers. Depending on your settings, it may be saved in your history and used to improve future models.
Never paste: passwords, PINs or security codes, your Social Security number, full bank or card numbers, medical records, other people's private details, or confidential work documents your employer hasn't approved. Swap in placeholders like “[my address]” or “Client A” instead.
| App | Stop chats being used for training |
|---|---|
| ChatGPT | Settings › Data controls › turn off “Improve the model for everyone.” Or use a Temporary Chat, which isn't saved to history or used for training. |
| Claude | Settings › Privacy › turn off “Help improve our AI models.” |
| Gemini | Settings & help › Activity › turn off Keep Activity. Google still keeps chats up to 72 hours to run the service. |
Work and school accounts often have different rules set by your employer or school. Check their AI policy before using a chatbot with work files.
FAQCommon questions
Do I need to say please and thank you?
It isn't required, and it won't hurt. What matters far more is clear instructions and good context. Writing the way you'd talk to a helpful person tends to produce clear prompts anyway.
Are long prompts better than short ones?
Not automatically. A prompt should be as long as it needs to be. “What's the capital of Peru?” is perfect as is. A cover letter request needs your job, experience and the job posting. Add detail when the chatbot would otherwise have to guess.
Does the same prompt work in ChatGPT, Claude and Gemini?
Yes. The same habits work in all of them. You may get different styles of answer, so if one result isn't what you want, follow up or try the prompt in another app.
Why did I get a different answer when I asked again?
Chatbots add a little randomness when choosing words, so the same prompt can give different answers. That's handy for ideas. For facts, a different answer each time is a hint to double-check.
What are “custom instructions”?
Many apps let you save background about yourself, like your job or how you like answers formatted, so you don't have to repeat it in every chat. ChatGPT calls this personalization or custom instructions, and Claude and Gemini have similar options. Don't save anything there you wouldn't type into a chat.
Can I use AI to write things for school or work?
It depends on the rules where you are. Many schools and employers allow AI for brainstorming but not for final work, or require you to say when you used it. Check the policy, and always read and fact-check anything before you hand it in.