Most people use AI the same way every day. They open a chat window, type a question the way they would text a friend, read whatever comes back, and quietly conclude that the tool is overrated.
The tool is rarely the problem. The instruction is.
Prompt engineering sounds like a technical skill reserved for developers. It is not. It is simply the habit of telling a machine clearly what you want, in a form it can act on. If you can brief a colleague, you can write a good prompt.
What actually separates a weak prompt from a strong one
Compare these two.
Weak: Write about digital marketing.
Strong: You are a marketing consultant advising a small clothing brand in Punjab with a budget of ten thousand rupees a month and no existing following. Write a one-page plan for their first ninety days. Give me five specific actions, ordered by what to do first, with a realistic result for each. Use plain language, no jargon.
The first prompt gets you a Wikipedia summary. The second gets you something you can act on tomorrow morning. Nothing changed about the AI between those two requests. Everything changed about the instruction.
The four parts of a reliable prompt
Almost every prompt that works well contains four things. Once you can name them, you stop guessing.
1. Role
Tell it who to be. “You are an experienced school teacher.” “You are a code reviewer.” This is not a magic trick, it is a filter. It tells the model which vocabulary, which assumptions and which level of detail to reach for.
2. Context
Give it the facts of your situation. Who is this for, what have you already tried, what are your real constraints. This is the part most people skip, and it is the part that does the most work. An AI cannot guess that your budget is small or that your audience is beginners unless you say so.
3. Constraint
Set the boundaries. Word count, tone, what to avoid, what must be included. Without constraints you get the average of everything the model has read, which is exactly what “generic” means.
4. Format
Say how you want it delivered. A table. Five bullet points. A short email. A step-by-step list. If you do not specify a shape, you will get paragraphs, and paragraphs are rarely what you actually needed.
Three mistakes that waste the most time
- Asking one enormous question. Break big tasks into steps. Ask for an outline first, then expand each section. You get better output and you can correct course early.
- Accepting the first answer. The first response is a draft, not a verdict. “Make this shorter.” “That example does not fit my situation, try another.” Treat it as a conversation.
- Not saying what you did not like. “Rewrite it” gives you a different version of the same problem. “Rewrite it — the tone is too formal and the second point is obvious” gives you a fix.
How to practise this week
Pick one task you already do regularly — writing emails, planning lessons, summarising documents. Write your normal prompt. Then rewrite it with all four parts: role, context, constraint, format. Run both. Keep the better one.
Do that for a week on the same task and you will have a prompt you can reuse forever. That is the whole discipline. Not memorising tricks, but building a small library of instructions that work for the specific work you do.
If you would rather not build that library from scratch, my AI Prompt Handbook contains over 300 ready-to-use prompts organised by task, plus the seven frameworks behind them. It works with ChatGPT, Claude, Gemini, Copilot and Perplexity.