O OpenChatGPT Start chatting

How to write better prompts

A prompt is the message you send to the AI. The better the prompt, the better the answer — most disappointing replies come from vague questions, not from a weak model. These habits will improve almost every answer you get.

Advertisement

Why the prompt matters

An AI chat model has no memory of your life, your project, or your last conversation — it only knows what is in your message and the last few messages of the current chat. When you write "Write an email", the model has to guess who the email is for, what it is about, and what tone to use. When you write "Write a short, polite email to a client asking for a one-week extension on a website project, because the design revisions took longer than expected", there is nothing left to guess — and the draft comes back usable. Specificity is the single biggest lever you have.

1. Be specific

Replace vague words with concrete ones. Instead of "Tell me about dogs", try "What are the three most common health problems in Labradors, and how can an owner spot them early?" Instead of "Help me with my essay", try "Help me outline a 1,000-word essay arguing that public libraries should stay open on Sundays, with three main points and one counter-argument." The model cannot read your mind, but it is very good at following a clear brief.

2. Give context

Tell the model who the answer is for and what it is for. "Explain photosynthesis to a 12-year-old" gets a completely different answer from "Explain photosynthesis to a university biology student". Context also means constraints: your budget, your deadline, your tools, your audience. "Plan a birthday party for 20 people with a budget of 5,000 rupees, no alcohol, in a small flat" is a question the model can actually solve; "Plan a party" is not.

3. Ask for a format

If you want a list, say so. If you want a table, a step-by-step guide, a short paragraph, or a 150-word product description, say so. Format instructions are the easiest way to get an answer you can paste straight into your work: "Give me 5 bullet points", "Make a table comparing the two options", "Write it as a formal letter", "Summarise this in one paragraph". Without a format, the model picks one for you — and it may not be the one you needed.

4. One thing at a time

Long, multi-part questions get muddled answers. If your message has five questions in it, the model may answer three and skip two, or mix the answers together. Break the work up: ask the first question, read the reply, then build on it with a follow-up. Chatting is a conversation, not a form — you do not have to fit everything into one message.

5. Iterate — treat the first answer as a draft

Almost no prompt is perfect on the first try, and that is normal. The fastest way to a great answer is a short back-and-forth: "Make it shorter", "Use simpler words", "Add an example", "That date is wrong — it was 1947, not 1952". Each correction narrows the result down. If a reply stops mid-sentence, just say "continue" and the model picks up where it stopped.

Prompt examples that work

Common mistakes

Before and after: the same question, rewritten

The fastest way to see what a good prompt does is to compare a vague question with a rewritten version of the same question:

Notice the pattern: every rewrite adds a goal, an audience, and a format. None of them needs special skill — just a few more words in the question.

Give the model a role

Starting a prompt with a role changes the whole answer. "Act as a patient maths teacher and explain fractions to a 10-year-old" gives a different reply from "Explain fractions" — the role sets the level, the tone, and the style in one line. Useful roles include a teacher, an editor, a career adviser, a code reviewer, and an interviewer. For example: "Act as a hiring manager and ask me five interview questions for a junior web developer job, one at a time, and give feedback on each answer before the next question." The role does not make the model smarter, but it makes the answer fit your situation instead of being generic.

Control length and tone

Length and tone are the two easiest things to control, and the two most often forgotten. Say the length in plain words: "in 5 bullet points", "in about 150 words", "in one paragraph", "in a table with three columns". Say the tone the same way: "formal", "friendly", "simple words a child can understand", "professional but not stiff". If you do not say, the model picks a middle length and a neutral tone — which is rarely what you wanted. And if the first answer is the right content in the wrong shape, a one-line follow-up fixes it: "Now make it half as long" or "Now make it sound more formal."

Ask for what you want, not what you don't want

Models follow positive instructions far better than negative ones. "Write a short summary" works better than "Don't make it too long". "Use simple words" works better than "Don't use difficult words". When you do need a limit, phrase it as a positive: "Keep it under 100 words" instead of "Don't write too much". The same applies to style: "Write it as a formal letter" beats "Don't make it sound casual". Tell the model the target, not the things to avoid.

Save your best prompts

When a prompt works well, keep it. A short note on your phone or a text file with your ten best prompts — one for emails, one for cover letters, one for debugging, one for study plans — saves you from rewriting the same question every time. Even better, save the whole pattern: "Write a [type of text] for [audience], about [topic], in [length], [tone]." Then you only fill in the blanks next time. Good prompting is mostly habit: the same five or six patterns cover most of what people ask an AI chat to do.

Short prompts beat long prompts

A common beginner mistake is writing a very long prompt, hoping more words will mean a better answer. Usually the opposite is true: a short, clear prompt with one task, one audience, and one format beats a paragraph of instructions. If your prompt is longer than three sentences, try cutting it in half and see whether the answer gets better or worse. You can always add detail in a follow-up — "now add an example" or "now make it formal" — once you have a first draft to work with. Short first, specific second, is the rhythm that works best.

A simple formula

When you are stuck, use this shape: task + context + format. "Write a cover letter (task) for a junior web developer job at a small agency, I have one year of freelance experience (context), in three short paragraphs, professional but friendly (format)." That one sentence carries everything the model needs. Try it on the chat page — and if you want the full beginner walkthrough, read how to use OpenChatGPT or the FAQ.