Prompting ChatGPT well comes down to four things: tell it who it is, give it the context, say what format you want, and show one example. Most weak answers are not the model failing; they are a vague request. If you ask "write me an email," you get a generic email. If you say "write a three-sentence email to a client who missed a payment, polite but firm, no apology," you get something you can actually send. This guide gives you the structure, a reusable template, and the habits that consistently raise answer quality.
The four-part structure that works
Almost every strong prompt has the same skeleton, whether you write it out or not.
- Role — a short framing of perspective, like "You are an editor reviewing a resume." Keep it to one line; it sets tone and assumptions.
- Context — the facts the model cannot guess: your audience, the goal, constraints, and any source text. This is where most of the quality comes from.
- Task — the specific thing to produce, stated as a verb: summarize, rewrite, compare, draft, critique.
- Format — the shape of the output: a table, a numbered list, a word count, a tone. Without it, you get a default essay.
You do not need fancy wording. "You are a careful proofreader. Here is a paragraph. Fix grammar only, keep my voice, return just the corrected text" outperforms any clever phrasing because every word does a job.
A reusable template
Paste this and fill the blanks. It works for writing, analysis, planning, and code.
| Slot |
What to put |
Example |
| Role |
A relevant perspective |
"You are a hiring manager." |
| Context |
Audience, goal, constraints |
"Reviewing a junior dev resume for a startup." |
| Task |
One clear verb + object |
"List the three weakest lines." |
| Format |
Shape and length |
"A table: line, problem, fix." |
| Example |
One sample of the result |
"Like: 'Managed projects' to 'Shipped 4 releases.'" |
The example row is the highest-leverage part and the one people skip. One sample of the output you want teaches style, length, and detail in a way instructions cannot. This is sometimes called few-shot prompting, and it works because the model pattern-matches your example.
How to iterate without starting over
The first answer is a draft, not a verdict. Stay in the same conversation and steer it:
- "Cut this in half and keep only the concrete points."
- "Too formal. Rewrite it the way you would text a coworker."
- "You invented a statistic. Remove anything you cannot support."
- "Give me three more options in the same format."
Refining beats rewriting because the model keeps the context you already built. If it drifts badly, paste the best version back and say "continue from this." For grounding it in real source material rather than its training memory, see what RAG is and when to use it.
What to skip
- Skip the giant persona preamble. "Act as a world-renowned, award-winning expert with 30 years of experience" adds words, not constraints. A plain role line is enough.
- Skip vague verbs. "Improve this" gives the model no target. Say what improved means: shorter, clearer, more formal.
- Skip trusting facts and numbers. ChatGPT can state confident, wrong figures. Verify anything load-bearing before you use it.
- Skip one giant prompt for a big task. Break it into steps and build on each answer; you keep control and catch errors early.
FAQ
Do I need to say "please" or use magic words?
No. Politeness does not change quality. Clear context, a specific task, and a stated format do. Spend your words there.
Why does ChatGPT ignore part of my instructions?
Usually the prompt is overloaded or contradictory. Put the most important constraint first, keep instructions short, and ask for one output shape at a time.
Should I write long prompts or short ones?
As long as needed for context, no longer. Include the facts the model cannot guess and the format you want, then stop. Filler dilutes the signal.
How do I get a consistent style every time?
Show one example of the result you want and say "match this style." For repeated tasks, save the working prompt and reuse it as a template.
Where to go next
Use ChatGPT for everyday work tasks, understand how ChatGPT actually works, and learn few-shot prompting in depth.