How to Write Better ChatGPT Prompts (With Examples)

Okay so here's the thing. When I first started using ChatGPT, I thought the tool was kind of underwhelming. I'd type something like "write me a blog post about productivity" and get back this bland, generic wall of text that sounded like it was written by a bored robot who had never actually struggled to finish a to-do list in their life. I figured that was just... what AI was. Mediocre by default.

I was wrong. The tool was fine. My prompts were terrible.

It took me probably three months of daily use before I started understanding what actually makes a prompt work. And once I got it, I genuinely couldn't believe the difference. Same tool. Completely different outputs. So I'm going to tell you exactly what shifted for me, with real examples, because the vague advice I kept reading online was not helping me at all.

The biggest mistake: being too vague

Most people write prompts the way they'd ask a question to Google. Short. Keyword-heavy. Minimal context. That works for search engines because search engines are matching you to existing documents. ChatGPT isn't doing that, it's generating something new based on what you give it. If you give it nothing, it fills in the blanks with whatever average, generic version of that thing exists in its training data.

So "write me a blog post about productivity" gets you a generic productivity post. Of course it does. You told it nothing about who you are, who's reading this, what angle you want to take, what you actually think about productivity, or how long it should be.

Compare that to: "Write a blog post about why to-do lists make people less productive, not more. The audience is people in their late 20s who feel overwhelmed by their task management systems. Tone should be conversational, a little irreverent, like a smart friend talking to you. Around 800 words. I want to argue that the problem isn't discipline, it's that most to-do lists don't account for energy levels."

That second prompt? You'll get something actually usable. Maybe not perfect, but close enough that editing it takes twenty minutes instead of starting over.

Give it a role. But make the role specific

You've probably heard "tell ChatGPT to act as an expert" before. And yeah, that works, but I've found that generic expert roles don't do much. "Act as a marketing expert" is still pretty vague. What kind of marketing? What era? What industry? What's their personality?

The roles that have worked best for me are weirdly specific. Like: "You're a copywriter who has worked in direct-response email marketing for SaaS companies for ten years. You write punchy, clear copy that converts. You hate jargon and always cut to what the reader actually cares about."

That specificity does real work. It tells the model what knowledge base to draw from, what aesthetic to aim for, and what to avoid. I've used this for technical explanations, resume help, recipe ideas, it works in way more situations than just writing tasks.

Tell it what you don't want

This one took me forever to figure out and I wish someone had told me earlier. You can tell ChatGPT what to avoid, and it actually listens. This is huge.

If I'm asking for an email and I don't want it to start with "I hope this email finds you well". I just say that. "Do not start with pleasantries. Get right to the point." If I'm writing something casual and I don't want it to sound stiff and formal, I say "don't use formal language or corporate-sounding phrases."

I keep a little mental list of my personal "never do this" instructions that I paste in when I need them. Things like: don't use bullet points, don't add a summary at the end, don't hedge everything with "it's important to note that." Every time the AI does something I hate, I add it to my don't-do list for that type of prompt.

Use examples, actual examples

If you want the output to sound a certain way, show it. This is probably the single highest-use thing I've found. Instead of describing a tone, paste in a paragraph that has the tone you want and say "write in a similar style to this."

I do this all the time for emails. I'll paste in an email I wrote that I liked and say "here's an example of how I naturally write. Match this voice for the email I'm about to describe." The difference between that and just saying "write in my voice" is enormous. You can't describe your voice, you just have it. Showing it is so much more effective.

Same goes for structure. If you want a specific format, show it. "Format this like the example below" and then give the example. Don't describe the format in abstract terms, just show the thing.

Iterate. Don't start over.

Okay real talk. I used to just regenerate when I didn't like an output. Hit the little regenerate button and hope for something different. That's not a strategy. That's chaos.

What actually works is telling it specifically what's wrong with what it gave you. "This is too formal, make it more casual." "The second paragraph is too long, break it up." "I liked the opening but the conclusion felt weak, rewrite just the conclusion to be more direct."

You can sculpt the output through conversation. That's the whole point of the chat interface. Use it. Each follow-up prompt is a chance to get closer to what you actually want instead of spinning out on full regenerations hoping to get lucky.

A quick template I actually use

Here's a rough structure I come back to when I'm writing a new type of prompt for the first time. It's not a magic formula. I don't think those exist, but it covers the things that matter most.

• Context: who am I, what am I working on, what's the situation
• Role: what kind of person/expert should ChatGPT be for this task
• Task: what exactly I want it to produce
• Tone/style: how it should sound, ideally with an example
• Constraints: length, format, things to avoid
• Goal: what this output is for, who will read it

I don't always include all of these for every prompt. Short tasks don't need all this. But for anything that matters, anything I'm actually going to use. I try to hit most of them.

The honest part

I'm not going to lie, even with all of this, sometimes the output still isn't great. AI has blind spots, and prompting is a skill that keeps developing the more you use it. But the gap between a thoughtful prompt and a lazy one is so wide that it's genuinely worth spending an extra two minutes before you hit send.

The people I've seen get the most out of ChatGPT treat it like a conversation with a really capable but context-blind collaborator. You have to give it the context it can't see. Once you start doing that consistently, the tool gets a lot less disappointing. And a lot more useful.