ChatGPT Prompt Engineering Basics: How to Write a Prompt That Gets the Answer First Time

The short version: a prompt gets it right first time when it tells ChatGPT who to be, what it already knows about your situation, what you want it to do, what shape the answer should take, and what to avoid. Skip any one of those five and you'll get something generic that you then have to rewrite anyway. Most people skip four of the five and blame the tool.

Why your first prompt usually misses

Here's what happens in most people's ChatGPT sessions. They type something like "write me a LinkedIn post about the new pension rules" and hit enter. ChatGPT gives back 200 words of confident, competent, utterly generic content that sounds like every other LinkedIn post about pension rules. Then they rewrite the prompt, add a bit more, get something slightly better, rewrite again, and by the fourth attempt they've got something usable. They call this "prompt engineering." What happened is they did in four goes what one well-built prompt would have done in one.

The tool isn't guessing badly. It's answering the question you asked, and the question you asked was too thin for it to do anything else. ChatGPT has no idea what your firm's tone is, who reads your LinkedIn, what "new pension rules" you mean, or what you're going to do with the post once it's written. So it fills the gaps with the most statistically average version of that content it can produce. That's not a flaw. That's exactly what a thin prompt earns you.

The five-part structure that gets it right the first time

I use the same five checkpoints for almost every prompt I write, whether it's for a blog post, a sales email, or a data summary. Miss two of these and you're back to editing.

• Role: who should ChatGPT be for this task. Not "an expert" (meaningless) but specific: "a UK-based financial adviser writing for other advisers, not consumers."
• Context: the facts it needs and can't guess. Your audience, your company, the specific thing you're referring to, any background it has no way of knowing.
• Task: the actual instruction, stated as a verb. Write, summarise, compare, rewrite, shorten, translate.
• Format: length, structure, tone, whether you want bullet points or prose, first person or third.
• Constraints: what to leave out. No jargon, no emojis, no exclamation marks, nothing that sounds like a press release, stay under 150 words.

That's it. Not a magic formula, not a trick phrase you slot in before your question, just five gaps you have to fill so the model has enough to work with. If you want a stash of ready-made starting prompts to practise this structure on, I've put together 40 simple ChatGPT prompts for beginners that already have most of these five parts built in, so you can see the pattern before you write your own.

A real example from a client session

Last year I was working with a financial services client, a small advisory firm in Manchester with about eleven staff, on getting their marketing team using ChatGPT for content instead of outsourcing every post to a freelancer. Their first attempt at a prompt was exactly the one-liner I described above: "write a LinkedIn post about the pension changes coming in April." The output was fine. It was also indistinguishable from what a hundred other advisory firms would post that week, and it used the word "unprecedented," which their compliance team hated on sight.

We rebuilt it together, live, in about six minutes. Here's what the second version looked like:

• Role: "You are a financial adviser writing directly to other advisers and high-net-worth clients, not the general public."
• Context: "The pension changes coming into effect this April affect annual allowance tapering for people earning over 260,000 pounds. Our audience is mostly business owners aged 45 to 65 who already have pensions and want to know if this affects them, not people learning about pensions for the first time."
• Task: "Write a LinkedIn post explaining what's changing and one practical question they should ask their adviser before April."
• Format: "150 words maximum, first person, plain English, no bullet points, ends with a question, not a call to action."
• Constraints: "No jargon like 'unprecedented' or 'game-changing.' Nothing that sounds like a press release. Don't use exclamation marks."

The output on that second try went into their content calendar unedited. Not because the prompt was clever, but because it contained everything the person writing it already knew in their head and just hadn't typed out. That's the whole trick, and it's a bit deflating to say it plainly: there's no shortcut past knowing your own answer well enough to describe it.

The uncomfortable bit nobody selling prompt courses wants to say

Here's the part that sits badly with a lot of "master prompt engineering" content out there. Most of the phrases people teach as prompt engineering tricks, "act as an expert," "think step by step," "you are a world-class copywriter," do very little on their own with current models. They occasionally nudge tone, but they don't compensate for missing information. I've tested this directly: the same thin request with "act as a world-class marketing strategist" bolted on the front produced content barely different from the version without it. What moved the needle every single time was adding real context, specific audience detail, and a clear format instruction.

So the uncomfortable truth is this: prompt engineering isn't really a technical skill you learn from a cheat sheet. It's clear thinking about your own task, typed out in full, instead of left in your head where ChatGPT can't see it. People who write vague briefs to human freelancers also write vague prompts to ChatGPT, and the tool exposes that habit faster than a human copywriter ever would, because a human will ask you follow-up questions and ChatGPT usually just guesses and runs with it.

A simple three-pass method if the five-part structure feels like a lot to hold in your head

If writing all five elements in one go feels clunky at first, build the prompt in layers instead.

1. Pass one, brain dump. Write everything you know about the task in messy sentences, no structure needed. Who it's for, what's happened, what you want, roughly how long.
2. Pass two, sort it. Reread what you wrote and label each sentence: is this role, context, task, format, or constraint. You'll usually find you've written plenty of context and task but forgotten format and constraints entirely, that's the normal gap.
3. Pass three, tidy and send. Reorder so role comes first, context second, task third, format and constraints last. Send that.

This takes maybe ninety seconds longer than firing off a one-liner. It saves you the three or four follow-up prompts you'd otherwise send trying to drag the answer into shape, and if you're watching your daily usage, that matters more than people think. I've written up the actual daily prompt limits for free and paid ChatGPT accounts and one thing is clear: most people who hit the limit aren't asking too much of the tool, they're asking the same thin question five times in a row because the first four answers weren't specific enough to use.

Where this goes wrong even for people who know the structure

Knowing the five parts doesn't automatically save you, because there are three habits that quietly sabotage even a well-built prompt.

• Burying the actual task in the middle of a paragraph. ChatGPT weighs the whole prompt but people read it in order, and so, functionally, does the model's attention on longer prompts. Put your task as its own short sentence, not tucked inside three sentences of throat-clearing.
• Giving format instructions that contradict themselves. "Make it punchy and conversational but also comprehensive and formal" is two different briefs. Pick one.
• Assuming one prompt has to do everything. If you need a blog post, a LinkedIn version, and three tweet variants, that's three prompts, or one prompt asking for all three explicitly and separately labelled. Don't ask for "content about X" and expect it to guess the format hierarchy you had in mind.

What this looks like for marketing tasks specifically

Marketing prompts fail in a particular way: people ask for tone without giving examples of the tone. "Write it in our brand voice" means nothing to a model that's never seen your brand voice. Paste in two paragraphs of your best existing content as a reference alongside the instruction, and the output shifts noticeably. I've collected a set of prompts built specifically around this problem, with the reference material built into the prompt itself, in AI prompts for marketers that work, and the pattern is the same five-part structure every time, just with a sample paragraph added as part of the context.

Building the habit

If you want to get past treating this as a one-off trick and build prompt writing into how you work day to day, it's worth treating it as a skill with a learning curve rather than a hack you either know or don't. I go through this in more depth, including how long it realistically takes to get fast at it, in how to learn prompt engineering basics for AI tools. The short version of that longer piece: give yourself two weeks of deliberately writing the five parts out in full, even when it feels slow, before you let yourself shortcut it. After two weeks it stops being a checklist and starts being how you naturally phrase a request.

One more thing worth knowing

Models update, and the way they handle thin prompts changes slightly with each version, sometimes getting better at inferring what you meant, sometimes not. What doesn't change is the underlying logic: more relevant, specific detail produces a more usable answer, and there's no version of ChatGPT coming that rewards vagueness. So the five-part habit is one of the few AI skills that won't go stale on you in six months.

I built a free prompt builder for exactly this.

Related guides live in the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.

I covered the follow-on problem in Mistakes Most First-Time Business Owners Make and How to Avoid Them.

Frequently asked questions

What's the single biggest mistake people make when writing ChatGPT prompts?

Asking a one-line question and expecting the model to fill in your audience, tone, and purpose on its own. It will guess, and the guess is always generic. Adding two or three sentences of context before the actual task removes most of that guesswork.

Do special phrases like "act as an expert" improve ChatGPT's answers?

Not much on their own. They can nudge tone slightly, but they don't make up for missing context, format instructions, or constraints. Specific audience and situation detail moves the answer far more than role-play phrasing does.

How long should a good ChatGPT prompt be?

There's no fixed length, but a prompt that covers role, context, task, format, and constraints usually runs three to six sentences for most business tasks. Shorter than that and you're likely missing at least one of those five elements.

Is prompt engineering something you can learn quickly, or does it take practice?

The structure itself takes about ten minutes to learn. Getting fast and natural at applying it without a checklist takes closer to two weeks of regular use, mainly because it forces you to get clear on your own brief, which is the actual skill being practised.

Sources worth reading

• OpenAI's own platform documentation