How to Write Better Prompts in ChatGPT

When using ChatGPT, the vast majority of people just type a short question, get a mediocre answer back, and conclude "AI isn't that good yet." That's not really what's happening though. It's more like they're giving directions to a very capable assistant who has zero context on them, and then getting annoyed when the assistant guesses wrong.

The fix isn't learning some secret hacker syntax. It's just being a clearer communicator. Here's how.

Why your prompts aren't working

Try this test: imagine you hired a smart freelancer, but they've never met you, don't know your business, and can't ask follow-up questions. If you just told them "write me a marketing email," you'd get something generic back. Not because they're bad at their job. It's because you gave them almost nothing to work with.

That's what's happening every time you type a one-line request into ChatGPT. The model isn't reading your mind. It's filling in every gap you left with its best generic guess, and generic guesses make generic output.

The good news: closing those gaps takes maybe 20 extra seconds of typing, and the difference in output quality is huge.

Give it a role

This one sounds like a gimmick but it genuinely changes the tone and depth of what you get back. Telling ChatGPT who to be before it answers narrows down which "version" of its knowledge it pulls from.

"You're a career coach who specializes in helping people switch industries. I want to move from teaching into UX design. What should my first three steps be?"

Compare that to just asking "how do I get into UX design." Same question, wildly different depth of answer, because now it's not just listing Google-able facts. It's answering as someone who does this for a living.

Tell it what the output should look like

If you don't say how you want the answer formatted, you'll usually get a wall of text. Say it up front instead.

"Give me this as a short bulleted list, no more than 5 points, written for someone who's never used a spreadsheet before."

That last part matters too. "Written for someone who's never used a spreadsheet before" is doing a lot of work. It tells ChatGPT what to leave out, not just what to include.

Give it context it doesn't have

ChatGPT doesn't know your business, your audience, your deadline, or what you already tried. If those things matter to the answer, say them.

Bad:

"Write a LinkedIn post about our new product."

Better:

"Write a LinkedIn post announcing our new product, a budgeting app for freelancers. Our audience is mostly self-employed creatives who are bad at saving for taxes. Keep it under 150 words, casual tone, no corporate buzzwords like 'synergy' or 'game-changer.'"

The second one gives ChatGPT actual material to work with instead of forcing it to invent your product, your audience, and your tone from nothing.

Break big asks into smaller ones

"Write my business plan" is a task most humans would need a week and several conversations to do well. Asking ChatGPT to do it in one shot usually gets you something shallow and generic, because it's trying to cover 10 sections at once instead of doing one well.

Instead, work through it in stages:

"Let's start with just the summary section of my business plan. It's a mobile dog-grooming service in a mid-sized city. Ask me any questions you need answered before you write it."

Then move to the next section once that one's solid. This takes a bit longer but the output is night and day better, and you're not stuck fixing one giant messy draft afterward.

Show it an example

If you have any idea what "good" looks like, show it. ChatGPT is very good at pattern-matching to an example, even a rough one.

"Here's an email I wrote for a similar situation last year: [paste it]. Write a new one in that same style, but for a client who's 3 weeks late on payment instead of 1 week."

This works especially well for tone. It's far easier to show ChatGPT what "sounds like me" looks like than to describe it in adjectives.

Ask it to ask you questions first

This one surprises people. You can literally tell ChatGPT to interview you before it answers, instead of guessing at what you meant.

"I want you to write a fundraising email for my nonprofit, but before you do, ask me whatever questions you need to write it well."

It'll usually come back with 3-5 genuinely useful questions (who's the audience, what's the ask, what's the deadline) and the answer you get afterward is built on real information instead of assumptions.

Correct it instead of starting over

If the first answer isn't right, don't delete it and rewrite the whole prompt from scratch. Just tell it what's wrong, the same way you'd give feedback to a person.

"This is close, but it's too formal. Make it sound like I'm texting a friend, not writing a report."

ChatGPT remembers the conversation, so it'll adjust the existing draft rather than starting over blind. This is usually faster than trying to cram every requirement into one perfect prompt the first time.

Before and after

A few more side-by-side examples, since these tend to make the idea click faster than any explanation:

Planning a trip

• Before: "Plan me a trip to Italy."
• After: "Plan a 7-day Italy trip for two people who like food and history more than nightlife. Budget is mid-range, not backpacker, not luxury. We're flying into Rome and want to end somewhere we can fly home from directly. Give me a day-by-day outline, not a full itinerary yet."

Learning something new

• Before: "Explain machine learning."
• After: "Explain machine learning to me like I'm a small business owner with no technical background. I want to understand it well enough to know when a vendor is overselling me on 'AI-powered' features."

Writing an email

• Before: "Write an email declining a meeting."
• After: "Write a short, polite email declining a meeting invite from a client. I want to keep the relationship warm. I'm just genuinely too busy this week. Suggest an alternative time next week instead of just saying no."

Getting feedback on writing

• Before: "Is this good?" (paste text)
• After: "Review this cover letter as if you're a hiring manager who reads 100 of these a week. Tell me the 3 biggest reasons you'd skim past it."

A prompt template you can reuse

If you want something simple to fall back on, fill in these blanks:

"Act as a [role]. I need [specific task]. The context is [relevant background info]. Format it as [format]. The tone should be [tone]. Before you answer, ask me anything you need clarified."

You won't need all five parts every time, but even using two or three of them will put you ahead of 90% of prompts people type in.

None of this is really about "tricking" ChatGPT into being smarter. It's about giving it the same information you'd give a person before asking them to do something for you. Once that clicks, prompting mostly stops feeling like a skill and starts feeling like just... talking clearly.