Most people type a vague request into ChatGPT, get a bland answer, and conclude AI isn’t that useful. The problem is usually the prompt, not the tool. This guide gives you a repeatable four-part formula (role, task and context, format, constraints), ten prompts you can paste in right now, and the small upgrades that make every future prompt better.
Here’s something I see in almost every workshop. Someone opens ChatGPT, types “write me a marketing email,” reads the generic result, and quietly decides the hype was overblown. The tool didn’t fail them. The instruction did.
ChatGPT now has around 900 million weekly users, and writing is one of the single most common things they ask it to do. [1] [2] So the skill of asking well isn’t a niche technical thing anymore. It’s closer to knowing how to write a clear email: a basic professional ability that quietly separates people who get value from AI from people who give up on it.
Good news: you don’t need clever tricks or secret keywords. You need a structure. Once you have one, every prompt you write gets better, and you stop staring at a blank chat box wondering why the output feels so flat.
Why most prompts fall flat
A chatbot predicts the most likely useful response to what you typed. If what you typed is vague, the most likely response is also vague. Ask for “a social post about our new feature” and you’ll get something that could belong to any company on earth, because you gave it nothing that makes your company yours.
Think about how you’d brief a sharp new freelancer. You wouldn’t just say “write me something.” You’d tell them who it’s for, what you’re trying to achieve, what good looks like, and what to avoid. AI is the same. The model is capable. It’s just waiting for a real brief.
Stop thinking of prompts as search queries and start thinking of them as briefs. A search box wants three keywords. A brief wants context, intent, and a definition of done. The more you treat the chat box like a colleague you’re delegating to, the better your results get.
The 4-part prompt formula
Here’s the structure I teach first, because it covers about 80% of everyday tasks. Four parts, in plain English:
- Role. Tell the AI who to be. “You are an experienced HR manager,” “Act as a copy editor.” This sets the tone, vocabulary, and level of detail.
- Task and context. Say exactly what you want and give it the background it can’t guess: your audience, your goal, the situation.
- Format. Describe what the output should look like. A table, five bullet points, a 150-word email, three options to choose from.
- Constraints. Set the boundaries. Tone, length, what to avoid, reading level, words to never use.
You won’t always need all four. But when an answer disappoints you, it’s almost always because one of these was missing. Run through them like a checklist and you’ll spot the gap.
Walking through each part
Role does more than it looks. “Explain compound interest” and “You are a patient teacher explaining compound interest to a 15-year-old” produce very different answers. The role quietly controls assumptions about what you already know.
Task and context is where most of the value lives. The AI can’t see your inbox, your customers, or last quarter’s numbers. Anything that matters has to be in the prompt. If the output feels generic, this is usually the part you skipped.
Format saves you the most time. If you want something you can paste straight into a slide, ask for it that way. “Give me this as a 3-column table” or “write exactly 5 subject lines, no explanations” stops the AI from wrapping everything in three paragraphs of preamble.
Constraints are how you sound like you and not like a robot. “Keep it warm but not salesy, under 120 words, no exclamation marks, and never use the word ‘delighted'” gets you a draft you’ll actually send.
You are an experienced HR manager. Write a short message to my team announcing that we’re moving to hybrid working, three days in the office. Context: people are nervous about losing flexibility, and I want to sound reassuring, not corporate. Format: a single message under 150 words. Constraints: warm and direct, no jargon, no phrase like ‘exciting journey.’
10 prompts you can copy right now
Paste any of these, swap the bracketed bits for your details, and tweak from there. They all use the formula above, so you can see the pattern repeat.
- “You are a plain-English editor. Rewrite this paragraph so a busy executive can skim it in 10 seconds. Keep every fact. Cut filler. [paste text]”
- “Act as a skeptical customer. Here’s my product description. List the three objections you’d have before buying, and what would change your mind. [paste]”
- “You are my meeting prep assistant. Turn these messy notes into a one-page brief: objective, key decisions needed, and three questions I should ask. [paste]”
- “Summarise this report into five bullet points a non-expert would understand, then one sentence on why it matters to a small business. [paste]”
- “You are a careful proofreader. Find grammar and clarity issues in this email, show them in a table (issue, fix, why), and don’t change my voice. [paste]”
- “Give me three versions of this LinkedIn post: one direct, one story-led, one with a bold opinion. Each under 100 words, no hashtags. [paste]”
- “Act as an interviewer for a [job title] role. Ask me one question at a time, wait for my answer, then give brief feedback before the next.”
- “Explain [topic] to me like I’m smart but completely new to it. Use one everyday analogy and avoid acronyms. If you must use one, define it.”
- “You are a data assistant. Here’s a list of survey comments. Group them into themes, count how many fall in each, and quote one example per theme. [paste]”
- “Draft a polite reply declining this request. Keep the relationship warm, give a brief honest reason, and offer one alternative. Under 90 words. [paste]”
Every one names a role, gives real context, asks for a specific format, and sets a limit. That’s the formula doing the work. Once you see it, you’ll never go back to typing three vague words and hoping.
Three upgrades that compound
1. Give it permission to push back. Add “if anything is unclear, ask me before answering” and the AI will often surface the one detail that would have made the whole thing wrong. This single line prevents a lot of confident nonsense, a pattern worth understanding (we cover it in our guide to the prompt patterns that still work).
2. Show an example. Paste one piece of writing you like and say “match this tone.” Models are far better at copying a sample than interpreting an adjective. If you want consistency across a whole team, this is also the foundation of training AI on your brand voice.
3. Iterate, don’t restart. The first answer is a draft, not a verdict. “Make it shorter,” “more skeptical,” “now as bullet points” gets you there faster than rewriting the prompt from scratch. The conversation is the tool, not just the first message.
Mistakes that quietly ruin your results
Stacking ten tasks into one prompt. If you ask for research, a draft, a critique, and a rewrite all at once, you get a mediocre blend of all four. Do them in steps.
Trusting numbers and quotes without checking. Let’s be honest: AI will sometimes invent a statistic that sounds perfect. For anything factual, ask for sources and verify them yourself. The formula makes answers better, not infallible.
Forgetting the AI has no memory of your world. It doesn’t know your last campaign flopped or that your CEO hates the word “synergy” unless you tell it, or set it up once in a workspace (see how Claude Projects handles persistent context).
What to do this week
Pick one task you do often: a recurring email, a weekly update, a type of summary. Write it once using the four-part formula. Save that prompt somewhere you’ll find it again, a note or a doc. Next time the task comes up, you paste, swap a few details, and you’re done in a fraction of the time.
That’s the whole game. You’re not learning to code or memorising tricks. You’re building a small library of good briefs for the work you already do. Do that five times and prompting stops being a skill you think about. It just becomes how you work.
Frequently asked questions
What makes a good AI prompt?
A good prompt gives the AI four things: a role to play, a clear task with real context, the format you want the answer in, and any constraints like tone or length. Vague prompts produce vague answers, so the more specific your brief, the better the result.
Do I need different prompts for ChatGPT, Claude, and Gemini?
Not really. The four-part formula works across all of them because they all respond better to clear briefs than to keywords. You may notice small differences in tone or length, but the same well-structured prompt will perform well in any of the major chatbots.
Are there magic words that make AI work better?
No. Phrases like “you are an expert” help mainly because they set a role, which is one part of the formula. There’s no secret password. Clear context and a defined output beat any single keyword every time.
How long should a prompt be?
As long as it needs to be to remove guesswork, and no longer. A one-line prompt is fine for a simple task. For anything where the result matters, a short paragraph that covers role, task, format, and constraints will outperform a single sentence almost every time.
Why does the AI sometimes ignore part of my prompt?
Usually because the prompt asked for too many things at once or buried the key instruction in the middle. Break complex requests into steps, put the most important instruction first or last, and tell the AI explicitly what to prioritise.
About this guide
This is a practical, non-technical guide to writing AI prompts that actually work. It’s built around a simple four-part formula, ten copy-paste examples you can adapt today, and honest warnings about where prompting goes wrong. Usage figures are drawn from OpenAI and reporting on ChatGPT’s user numbers.
Related reading
Sana is an AI educator and learning designer specialising in making complex ideas stick for non-technical professionals. She has trained 2,000+ learners across corporate teams, bootcamps, and keynote stages. Future Factors offers AI Bootcamps, Corporate Workshops, and Speaking & Consulting for businesses ready to adopt AI without the overwhelm.
More about Sana →- [1] TechCrunch. ChatGPT reaches 900M weekly active users. 2026.
- [2] OpenAI. How people are using ChatGPT. 2025.
- [3] CNBC. OpenAI study revealing how people use ChatGPT. 2025.
- [4] McKinsey. The state of AI in 2025. 2025.
- [5] Microsoft WorkLab. Work Trend Index 2025. 2025.
