Most people’s first ChatGPT prompt is something like “write me an email” — and it shows. The reply is generic, a bit flat, and you spend longer editing it than you would have spent writing it yourself. The problem isn’t the tool. It’s that a vague instruction gets a vague answer.
The good news is that prompting is a learnable skill with a short ramp. You don’t need special syntax or a magic phrase. You need a structure, a few templates you can reuse, and a willingness to push back when the first answer is mediocre. Here’s what actually works.
The one habit that separates good prompts from bad ones
Before you type anything, ask yourself: could a competent freelancer do this task with only the information I’ve given? If the answer is no, ChatGPT will struggle too.
A useful prompt usually carries four things: what you want, who it’s for, how long or detailed it should be, and any constraints. “Write an email” becomes “Write a 120-word email to a client explaining that our delivery is delayed by five days, apologising without over-explaining, and offering a 10% discount on the next order.” Same tool, completely different output.
If you’re still working out where ChatGPT fits into your day, a solid starting point is a beginner’s guide to what to try first in ChatGPT — it covers the tasks worth delegating and the ones you should keep doing yourself.
Beginner prompts for writing and editing
The rough-notes-to-draft prompt
Don’t ask ChatGPT to write from nothing. Give it your messy bullet points and let it do the structuring.
“Here are my rough notes from a client call: [paste notes]. Turn these into a follow-up email with three short paragraphs — what we agreed, what I need from them, and next steps with dates. Keep the tone friendly but professional, no filler phrases.”
This works because you’re supplying the substance. ChatGPT handles the tidying, which is where most of the time goes.
The “cut 20%” prompt
Almost every first draft is too long. Instead of asking for a rewrite, be specific about the target:
“Cut this by about 20% without losing any facts. Remove hedging words like ‘really’, ‘just’ and ‘very’, and keep my sentence rhythm — don’t make it sound like a press release.”
That last clause matters. Without it, you’ll get something technically shorter but weirdly corporate.
Prompts for planning and decision-making
ChatGPT is genuinely useful as a thinking partner, but only if you stop it from agreeing with you. Left alone, it will validate whatever plan you bring. Force it into the other chair.
- Steelman the opposite: “Here’s my plan for [X]. Argue against it as a sceptical colleague would. List the three weakest assumptions.”
- Generate options, not answers: “Give me five different ways to structure a two-week launch, ranging from cheapest to most aggressive. One line of reasoning each.”
- Pre-mortem: “It’s six months from now and this project failed. Write the five most likely reasons.”
- Explain the trade-off: “Compare [option A] and [option B] on cost, speed and risk. Use a short table, then tell me which you’d pick if speed mattered most.”
These prompts produce material you’ll actually argue with, which is the point. If you want to go further into building tools rather than just chatting, there’s a walkthrough on building your first AI chatbot in an afternoon that assumes no coding background.
Prompts for learning something new
“Explain quantum computing” gets you a Wikipedia summary. You want a tutor that adapts, so build that into the prompt.
“Explain how compound interest works to someone who failed maths at school. Use one concrete example with real numbers — £2,000 saved at 5% over 10 years. Then ask me a question to check I understood.”
That final instruction changes the dynamic from lecture to conversation. A follow-up that works well: “Now explain it again, but as if I’m a 14-year-old who’s bored.” Different framing surfaces different explanations, and one of them will click.
For anything you want to learn by doing rather than reading, voice mode is underrated — you can talk through a problem out loud and get corrections in real time. It’s not flawless, though, and this breakdown of what ChatGPT voice mode handles well is worth reading before you rely on it for anything important.
Fixing a bad answer without starting over
Beginners often delete the prompt and try again from scratch. Don’t. The conversation has context now, so corrections are cheaper than restarts. Try these three moves in order:
1. Name the flaw. “Too formal. Rewrite it as if you’re emailing a colleague you like.”
2. Add a constraint. “Same content, but under 100 words and no bullet points.”
3. Ask for variants. “Give me three versions — one direct, one warm, one short enough for a text message.”
If two rounds of correction haven’t helped, the prompt is the problem, not the model. Rewrite the instruction from scratch with more specifics.
Three prompting mistakes worth avoiding
Asking for facts without checking them
ChatGPT will produce confident-sounding dates, statistics and citations that don’t exist. Treat every specific number as a claim to verify. Ask “what’s your source for that figure?” — if it can’t point to one, assume it made it up.
Letting it guess at missing context
If your prompt leaves a gap, ChatGPT fills it silently. Tell it what to do instead: “If anything in my request is ambiguous, ask me a clarifying question before answering.” This one line prevents a lot of wasted output.
Pasting confidential material
Client contracts, staff records, unpublished financials — none of it belongs in a chat window. Anonymise first: swap names for roles and round the numbers. The advice stays just as good.
Building a prompt library you’ll actually reuse
The real productivity gain arrives around week three, when you notice you’re typing the same instructions repeatedly. Keep a plain text file with your best ten prompts, and save the variables as placeholders — [audience], [word count], [tone].
Add to it whenever a prompt produces something you barely had to edit. That’s the signal it’s worth keeping. Ten refined prompts beat a hundred bookmarked ones you never open, and after a month you’ll find you write fewer, sharper instructions because you’ve internalised what the model needs to hear.
One last thing: pay attention to what you’re paying for. If you’re weighing up the paid tiers, this review of the ChatGPT Pro plan in daily use gives you a realistic picture of where the extra money shows up — and where it doesn’t.