How to Write ChatGPT Prompts That Actually Work

What is how to write ChatGPT prompts?

Writing a ChatGPT prompt means crafting the exact text you send to the model to get a specific, useful result. It sounds trivial — you type something, the model responds — but the gap between a prompt that produces a generic paragraph and one that produces a publication-ready output is enormous. A well-written prompt tells ChatGPT who it is, what you need, the context behind the request, any constraints on the output, and the format you want. A poorly written prompt gives it none of that and hopes for the best. Prompt writing is the skill of removing ambiguity from your request before the model has a chance to fill that ambiguity with the wrong assumptions.

Why most people fail at how to write ChatGPT prompts

Most people treat ChatGPT like a search engine: they type a short phrase and expect a complete, accurate answer. Search engines surface existing content for vague queries. ChatGPT is a prediction engine — it generates the most statistically probable continuation of whatever you give it. Give it a vague phrase and you get a vague continuation. The other failure mode is over-relying on the first response. ChatGPT's first output is a starting point, not a finished product. People who get bad results consistently write a single vague prompt, read the generic output, decide the tool doesn't work, and stop. People who get great results write one sentence, read the output, identify what's wrong, and add a targeted clarification. The difference is iteration mindset, not tool quality.

How to do how to write ChatGPT prompts properly

1. 1
   Assign a role before anything else

   Start with 'Act as a [specific expert]'. This single addition shifts which knowledge patterns ChatGPT draws on. 'Act as a senior Python developer with experience in data pipelines' produces different — and better — code than 'write me some Python code'. The role sets the register, depth, and vocabulary of the entire response.

2. 2
   State the context the model can't see

   ChatGPT knows nothing about you, your project, or your audience. Tell it. 'I'm writing for a technical audience that already knows REST APIs but has never used GraphQL' gives the model a filter for what to explain and what to skip. Without context, it writes for an imagined average reader who is probably not you.

3. 3
   Give the task as a clear imperative

   Use an action verb and a specific object: 'Write', 'Summarise', 'Rewrite', 'List', 'Compare'. Avoid vague framings like 'Tell me about X' or 'Help me with Y'. 'Summarise the three main risks of using microservices for a startup with a team under 10 people' is a task. 'Tell me about microservices' is a topic. Tasks get results; topics get overviews.

4. 4
   Add constraints to eliminate unwanted output

   Constraints are the most underused lever in prompt writing. Word limits, format restrictions, tone guidelines, and exclusions all narrow the solution space. 'Under 200 words', 'no bullet points — flowing prose only', 'do not mention pricing'. Each constraint you add is one less judgment call the model makes incorrectly.

5. 5
   Specify the output format explicitly

   ChatGPT defaults to a mix of prose and bullets that rarely matches your use case. Tell it: 'Format as a markdown table', 'Respond in JSON with keys: name, role, risk', 'Write three short paragraphs with no headers'. When you specify format, you eliminate the step of reformatting the output to fit your document or workflow.

6. 6
   Iterate instead of starting over

   If the first response is wrong, do not delete and retype. Add a follow-up: 'The tone is too casual — make it more formal while keeping the same structure' or 'The second paragraph is too vague — rewrite it with a specific example'. Targeted corrections in the same conversation outperform a fresh prompt 90% of the time, because ChatGPT retains the established context.

Examples — bad prompt vs good prompt

Write me an email about our product launch.

Act as a B2B SaaS copywriter. Write a 150-word launch announcement email for Kova, a project management tool for remote engineering teams. The audience is existing users on the free plan. Goal: get them to upgrade to Pro by highlighting the new async standup feature. Tone: conversational but not casual. End with a single CTA button labeled 'Unlock async standups'. No bullet points — flowing prose only.

Common mistakes

• ✗Writing one vague sentence and expecting a finished output — ChatGPT is a collaborator, not a mind reader.
• ✗Forgetting to specify tone — 'professional' and 'formal' mean different things, and neither is the default.
• ✗Not including the audience — content written for a beginner looks patronising to an expert and vice versa.
• ✗Asking multiple unrelated questions in one prompt — ChatGPT will answer all superficially or focus on the first and forget the rest.
• ✗Accepting the first output without iteration — the first response is a draft, not a deliverable.
• ✗Over-prompting on complex tasks without using multi-turn conversation — break large tasks into sequential steps.

Pro tips

• ✦Save your best prompts in a plain text file organised by task type. Reusing a refined prompt is always faster than rebuilding one from scratch.
• ✦Use 'before you answer, ask me any clarifying questions you need' when the task is complex — ChatGPT will identify gaps you hadn't thought of.
• ✦Add 'rate your own response out of 10 and explain where it fell short' at the end of your prompt. The self-critique is often more useful than the original output.
• ✦For code tasks, always append 'include error handling and comments explaining non-obvious logic' — this is never the default.
• ✦If the response is in the right direction but too long, say 'rewrite this at 60% of the current length without losing any key points' — more precise than 'make it shorter'.
• ✦Use numbered lists in your prompt when you have multiple requirements — ChatGPT is more likely to address all of them than when buried in a paragraph.

Continue learning