How to Write Prompts for ChatGPT: A Plain-English Guide for 2026

Hundreds of millions of people use ChatGPT every week, and most of them are leaving better answers on the table. The same question can return vague filler or something genuinely useful, and often the only thing that changed is how it was asked.

Most people type into ChatGPT the way they type into a search box: a few keywords, a half-formed question, then disappointment when the reply reads generic. The model usually isn't the problem. As of June 2026 the default, GPT-5.5 Instant, is sharp enough to draft an email or summarise a report, but it only works with what you give it. Give it a vague request and it fills the gaps with guesses.

The fix is a handful of small habits, none of which need code or jargon, starting with the one that solves most bad answers. If you've never opened ChatGPT at all, our how to use ChatGPT for beginners guide covers the basics first. If you want the bigger picture, what prompt engineering even is and why it works, our prompt engineering for beginners guide covers the concepts; this one is the hands-on how-to.

What a prompt actually is

A prompt is the instruction you give ChatGPT: the question, request, or task you type into the box. That is the entire input the model has to work from, so the quality of the prompt sets the ceiling on the quality of the answer.

Think of it less like a search query and more like briefing a fast, capable, slightly literal assistant who has no idea what is in your head. It knows a great deal in general, but it knows nothing about your situation unless you tell it. Everything you leave out, it fills in with an average guess.

The habit that fixes most bad answers: be specific

Vagueness in, vagueness out. The single biggest upgrade to your prompts is detail: who the answer is for, what it is about, and what you want back. Compare these two.

The vague version:

Write a marketing email.

That returns a generic template about a generic product, the kind of thing you delete on sight. Now the specific version:

You are an email marketer for a small coffee subscription business. Write a 120-word promotional email announcing a 20 percent discount for new subscribers this week. Use a warm, friendly tone, include one clear call to action, and suggest a subject line. Avoid jargon and exclamation marks.

That returns an email you can almost send as-is. Same model, same underlying task, completely different result, and the only thing that changed is the detail you handed over.

Tell it who to be

Giving ChatGPT a role focuses its answer. A role sets the perspective, vocabulary, and priorities it writes from, so "act as a financial advisor" and "act as a teacher" handle the same question very differently. It is shorthand for a whole set of instructions you would otherwise have to spell out.

Act as an experienced UX writer. Rewrite this button label to be clearer and more action-oriented, and give me three options with a one-line reason for each: 'Submit'.

The "act as" pattern is one of the oldest tricks in prompt writing, and it is popular because it works. A few words of role can lift an answer more than a whole paragraph of instructions.

Give it the context it can't see

ChatGPT cannot see your world. Context is the background it needs to be useful: your audience, your goal, what you have already tried, any constraints that matter. The more relevant context you add, the less it has to invent.

I run a two-person bakery and I'm writing our first Instagram post for a new sourdough loaf. Our customers are local, mostly families, and our tone is friendly and down to earth. Write three short caption options, each under 200 characters, with a question at the end to invite comments.

The answer now has a lot to work with: the business, the audience, the tone, the format, and the goal. That is the difference between a caption that sounds like you and one that sounds like it came off a conveyor belt.

Spell out the format you want

If you want a table, ask for a table. ChatGPT defaults to prose paragraphs unless you say otherwise, and a lot of frustration comes from getting an essay when you wanted a list. Name the shape of the output: a bulleted list, a table with specific columns, a short email, a step-by-step checklist, a word count.

Compare three budgeting methods (50/30/20, zero-based, and envelope) in a table. Columns: method, how it works in one sentence, who it suits, and the main drawback. Keep every cell under 15 words.

Format instructions also make the answer easier to reuse. A table drops straight into a document, and a checklist is ready to act on the moment you read it.

Show an example of what you want

When a style is hard to describe, show it. Giving ChatGPT one or two examples of the output you want, a technique often called few-shot prompting, is more reliable than describing the style in words. The model copies the pattern.

Turn these product features into benefit-led one-liners. Follow this example. Feature: 'Battery lasts 30 hours.' One-liner: 'Forget your charger for the whole week.' Now do these three: 1) Waterproof to 50 metres. 2) Noise-cancelling microphone. 3) Folds flat to fit a coat pocket.

Two good examples teach the model more than a paragraph of adjectives ever could. It is the technique professional copywriters and developers lean on most, because it shows rather than tells.

Set the tone, length, and limits

Constraints are where a good prompt gets its polish. Tell ChatGPT how long, how formal, what reading level, and what to avoid. Plain adjectives like formal, friendly, plain, or technical steer the tone, and a word count or a "no jargon" rule keeps it honest.

Explain how a credit score works to a 16-year-old who has never had a bank account. Keep it under 150 words, use one everyday analogy, avoid financial jargon, and don't use the word 'simply'.

A constraint worth keeping in your back pocket: ask for a specific reading level, or for "plain English a beginner could follow." It quietly changes the whole register of the answer.

Break big tasks into steps

One giant prompt often gets a shallow answer. For anything complex, you can get better results by splitting the job, or by asking ChatGPT to work through it in order. For reasoning-heavy tasks, asking it to think step by step before answering tends to surface fewer mistakes.

I want to plan a 3-day trip to Lisbon on a mid-range budget. First, ask me the three questions you most need answered before you can plan well. Wait for my replies before suggesting any itinerary.

Letting the model ask you questions first is an underused move. It turns a one-shot guess into a short back-and-forth that lands much closer to what you actually wanted.

Hand it your own draft

You don't have to start from a blank box. One of the most useful things ChatGPT does is improve text you already have, which sidesteps the generic-filler problem entirely, because the substance is yours. Paste your draft, say what you want changed, and keep your own voice in the room.

Tighten the rough cover-letter paragraph below. Fix clunky phrasing and keep my plain, direct tone, and don't add new claims about my experience. Return the edited version, then a one-line note on what you changed. Paragraph: [paste your paragraph].

Editing is lower-risk than generating, because you can see exactly what moved. It is also where the model is most reliable, since it works from your facts instead of inventing its own.

Ask it to pressure-test its answer

ChatGPT can critique its own work if you ask it to. Once it gives you something, you can turn it into its own editor: ask what is weak, what is missing, or what a sceptic would push back on. It often catches gaps it glossed over the first time.

Review the answer you just gave. What are its two weakest points, and what important objection or risk did you leave out? Then give me a tighter version that fixes them.

This works because the model is good at judging against a clear standard, even when its first attempt fell short of it. A round of self-critique is often faster than rewriting the prompt from scratch.

Treat it as a conversation, not a vending machine

Your first prompt is a draft, not a verdict. The real skill is iterating: read what comes back, then refine it. "Make it shorter." "More formal." "Give me three more like the second one." Each follow-up is a small new prompt, and ChatGPT keeps the thread, so you are steering rather than starting over.

This is why prompt writing is forgiving. You don't have to nail it in one go. You nudge the answer toward what you wanted, the way you would guide a keen assistant who needs a little direction.

A formula you can reuse

Once these habits click, most strong prompts follow the same shape. A reliable formula is Role, Task, Context, Format: who ChatGPT should be, what you want done, the background it needs, and the form the answer should take. You won't use every part every time, but it works as a checklist.

You are a [role]. [Task: what you want done.] Context: [audience, goal, constraints, anything relevant]. Give the answer as [format: a list, a table, an email, a set word count].

That skeleton, filled in, covers most everyday use. Keep it somewhere handy until it becomes second nature, and build your own collection of the prompts that work for you. We are putting together a free prompt library of ready-made ones to start from.

A worked example, start to finish

Watch the formula do its job on a real task. Say you need a polite reminder to a client who hasn't paid. The first instinct is something like "write a payment reminder email," which gets you a stiff, generic template. Fill in the formula instead:

You are a friendly but professional freelancer. Write a short reminder email to a client whose invoice is 10 days overdue. We have a good relationship, so keep it warm and assume they just forgot. Under 120 words, with a clear line stating the amount and a gentle next step, plus a subject line.

That gets you a usable email on the first try. If it lands slightly too soft, you refine: "a touch firmer, and add that this is the second reminder." Two short follow-ups and it is exactly right, which beats trying to nail the opening prompt in one shot. The formula gets you most of the way there, and the conversation closes the gap. When you want the email ones ready-made, our best ChatGPT prompts for email writing collects a dozen to copy.

Common mistakes to skip

A few habits quietly sabotage otherwise decent prompts.

• Asking for too much in one breath. Five requests in a single prompt usually returns five half-answers. Split them up.
• Leaving the format to chance. If you don't say how you want it, you get a wall of prose.
• Treating the first answer as final. The output is a starting point to refine, not the last word.
• Trusting facts without checking. ChatGPT can state a wrong date or a made-up source with total confidence, so verify anything that matters.

A note on which model you are using

Newer models forgive sloppier prompts, but the gap is smaller than the hype suggests. As of June 2026, ChatGPT's default is GPT-5.5 Instant, and OpenAI's own guidance notes that stronger models can reduce the need for elaborate prompting by 30 to 50 percent on instruction-following tasks. They still can't read your mind. The same prompt that works well in ChatGPT works, with small tweaks, in Gemini or Claude, because clarity travels across all of them. Model names and defaults shift every few months, so treat the specifics here as a snapshot of mid-2026.

What this post does not cover

• API settings like temperature and system messages, which belong to the developer tools, not the ChatGPT app
• Prompts for image, audio, or video models, which follow different rules
• Workarounds for the model's safety limits

Sources

1. OpenAI Help Center: Prompt engineering best practices for ChatGPT
2. OpenAI Academy: Prompting
3. OpenAI Help Center: Model release notes
4. IBM: The 2026 guide to prompt engineering