We searched "how to write ChatGPT prompts" last week and counted 47 articles in the first two pages of results. Nearly every one taught some version of the same formula: give the AI a role, define a task, specify a format, add constraints, list examples.
ROLE: You are a senior marketing strategist. TASK: Write a LinkedIn post about AI adoption. FORMAT: 3 paragraphs, professional tone. CONSTRAINTS: No hashtags. Under 200 words. EXAMPLES: Here are two posts I like...
That formula works. But it also turns a conversation into a configuration file, and it's the reason a lot of people feel like they're "bad at prompts." They think they need to memorize frameworks. They don't. They need to talk to ChatGPT the way they'd talk to a smart colleague who just joined their team and doesn't know anything about their project yet.
One principle. That's it: give context, then ask.
Just explain what you're working on and what you need, the same way you'd brief a person.
Bad prompt vs good prompt
Here's an example from our own work.
Bad: "Write a blog post about AI literacy."
This isn't bad because it's short. It's bad because the AI has no idea what you mean. "AI literacy" could be a policy brief, a beginner's guide, a critique, a syllabus, or a LinkedIn rant. The AI will pick one, and it probably won't be what you wanted.
Good: "I'm writing a blog post for professionals who are curious about AI but feel behind. The post should explain what AI literacy means in plain terms and give them 3 practical things they can do this week to improve theirs. Keep it under 1,500 words. Here's a post I wrote recently that shows the tone I want: [paste link or text]."
The difference isn't structure or formatting. It's context. The second version tells ChatGPT who the audience is, what the goal is, how long it should be, and what the tone sounds like. That's enough for the AI to produce something useful on the first try.
The one principle in practice
The principle is simple: give context, then ask. In practice, that means covering three things before you make your request.
What you're working on. One or two sentences about the project. "We're writing a blog post for professionals who feel behind on AI and need a starting point." "I'm preparing for a performance review." "I'm brainstorming names for a podcast." The AI doesn't know your life. Tell it.
Who the output is for. "My team is a mix of engineers and designers, and they're stressed about the timeline." "The audience is beginners who have never used AI." "It's for my manager, who likes concise updates." This shapes tone, depth, and framing.
What a good result looks like. "I want a short email that communicates urgency without panic." "Give me 10 options, some serious, some playful." "I need a structure I can react to, not a finished draft." This is where you set expectations for format and quality without turning it into a spec sheet.
Three things. No acronyms to remember.
The conversation loop
A common stuck point: you write one prompt, get a mediocre response, and conclude that either the AI is bad or your prompt was bad. Usually neither. The first response is a draft. You revise.
Think about how you'd work with a new colleague. You'd give them a brief, they'd come back with something, you'd give feedback, they'd revise. Same thing here. The first response is the starting point, not the finish line.
Response isn't what you wanted? Tell the AI what to change. "This is too formal. Make it conversational." "You used too much jargon. Simplify the language." "The structure is good but the examples are generic. Here are two to use."
Response is close but missing something? Add context. "This is good, but you're missing the part about how this connects to our Q4 goals." "Add a section for people who have never used AI at all."
Response is good and you want to go deeper? Ask a follow-up. "Now give me three variations of the opening paragraph." "What counterarguments should I anticipate?" "How would you shorten this by half?"
This loop is where the value is. Not in crafting the perfect single prompt. You probably won't nail it on the first try, and that's fine. The people who get great results from ChatGPT are the ones who iterate, not the ones who engineer.
We use this loop for every post on this blog. The first draft is never the one you read. The third or fourth version is. The same principle that makes a good prompt makes a good post: context, feedback, revision.
For a deeper look at this conversational approach, we wrote about the three behaviors that unlock most of ChatGPT's value: ask, loop, and give context.
When structure helps
We said no frameworks. That's slightly oversimplified. There are moments where adding structure to your prompt improves the result. The key is knowing when.
When you need a specific format. If you need a table, a numbered list, a JSON object, or a specific document structure, say so. "Format this as a table with columns for task, owner, and deadline." That's just telling someone what format you need.
When you're doing something technical. Code generation, data analysis, API interactions. These benefit from precise instructions. "Write a Python function that takes a list of URLs and returns their HTTP status codes. Handle timeouts gracefully." The more specific you are with technical tasks, the better the output.
When you want the AI to adopt a perspective. "Respond as if you're a skeptical CFO evaluating this proposal." "Explain this from the perspective of someone who has never used a smartphone." This isn't role-playing for its own sake. It's directing the AI's attention to a specific angle.
In each case, the structure serves the request. The request doesn't serve the structure. That's the difference between giving context and doing prompt engineering.
Common pitfalls
A few things tend to go wrong, and none of them require a formula to fix.
Too little context. "Write me a bio" gives you a generic bio. "Write me a bio for my company's about page. We're an AI education brand helping professionals catch up on technology. The tone should be warm and approachable, not corporate. Here are three facts to include" gives you something you can use. The fix is almost always the same: add more context.
Too much instruction. Long prompts with 15 constraints, 4 examples, and 3 formatting rules can produce worse output than short prompts with good context. The AI tries to satisfy every constraint and loses the thread. If your prompt is longer than the response you want, trim it.
Treating the first response as final. We covered this in the conversation loop, but it's worth repeating because it's the most common mistake. The first response is a draft. Revise it. The second or third version is almost always better.
Copying prompts from the internet. Prompt libraries are full of elaborate templates that someone else wrote for their specific use case. Pasting them into your conversation rarely helps because their context isn't your context. You're better off writing your own three sentences of context than copying someone's 200-word prompt template.
What about system prompts?
If you've gone deep into AI tools, you might have come across "system prompts," the instructions that configure an AI's behavior behind the scenes. System prompts matter for developers building AI applications. They don't matter for you if you're using ChatGPT in a browser window.
Your prompt is the message you type. That's what you control. Focus on making that clear and contextual. If you eventually start building custom GPTs or AI agents, you'll learn system prompts then. For now, they're not relevant to getting better results from a chat interface.
If you're curious about where this leads, our AI literacy guide covers the broader skill set, including when to trust AI output and when to verify it.
FAQ
Do I need to learn prompt engineering to use ChatGPT well?
No. Prompt engineering, as a formal discipline, is for developers who are building AI applications and need to control model behavior at scale. For everyday use, the principle is simpler: explain your context clearly and ask for what you need. The AI responds to clear communication, not to formulas.
What makes a ChatGPT prompt "good"?
A good prompt gives the AI enough context to understand what you want. That means explaining what you're working on, who the output is for, and what a good result looks like. The more specific and contextual you are, the better the response. Length doesn't matter. Clarity does.
Should I use prompt templates from online?
Use them as inspiration, not as copy-paste. A template someone wrote for their marketing workflow probably won't fit your situation. You'll get better results by writing a few sentences of your own context than by pasting in a 200-word template that was designed for someone else's use case.
How long should my prompt be?
As long as it needs to be and not longer. Some tasks need three sentences of context. Others need a paragraph. The goal is giving the AI what it needs to help you. If you find yourself writing a very long prompt, check whether you're adding context or just adding constraints. Context helps. Constraint lists usually don't.
Can I ask ChatGPT to improve my prompt?
Yes, and it's a useful trick. You can paste your draft prompt and ask: "I'm going to send you this prompt. Before I do, tell me what context is missing that would help you give me a better response." The AI will tell you what it needs. Then you add that and send the improved version.
Want to go further?
Pick something you're working on today. An email, a document, a brainstorm. Before you ask ChatGPT for help, write two sentences about what the project is and who the output is for. Then make your request.
Context, then ask. Everything else is iteration.
If you want a structured starting point for your AI journey, take the free GWT survey at greatwithtech.com. It takes 2 minutes and maps your AI readiness across practical skills.