By Prompt engineering is the skill of writing clear instructions so that an AI tool gives you useful answers instead of generic ones. A simple way to do it is a five-part checklist: Role, Task, Context, Format and Constraints. It works like giving a clear brief to a tutor who has never met you.
Why do vague prompts give generic answers?
Vague prompts give generic answers because the AI has to guess what you mean, and its safest guess is the most average one. Here is a common situation. A student types this into ChatGPT:
Out comes a long, textbook-style answer. It is correct, but it does not fit the student. It may be too advanced or too basic, with no example and no link to the exam. The student says, “This is generic. It does not help me.” They are half right. The answer is generic. But the tool did not fail. The instruction did.
Two terms will help. A large language model (LLM) is software trained on a huge amount of text to predict the most likely next words. ChatGPT, Claude and Gemini are built on LLMs. A prompt is the text you type to it. When your prompt is vague, the model has to cover every possible reader, so the safest answer is a general one. The model is not being lazy. It has nothing else to go on. New to this? Start with what generative AI is and how LLMs are trained.
The fix is not secret words or paid prompt packs. Think of asking a tutor who has never met you. If you say only “explain machine learning”, they give a standard lecture. If you say “I am a second-year commerce student, I know basic statistics, my exam is on Friday, explain it with a simple example”, they can really help. The framework below is that briefing, written as a checklist.
What are the five parts of a good prompt?
A good prompt has five parts: Role, Task, Context, Format and Constraints. Each one answers a question the AI cannot answer for itself.
| Part | Question it answers | Example fragment |
|---|---|---|
| Role | Who is answering? | “You are a patient statistics tutor…” |
| Task | What is the one job? | “…explain what a p-value means…” |
| Context | What does the AI need to know? | “…I am a second-year student. I know mean and standard deviation but not hypothesis testing. My exam is on Friday…” |
| Format | What should the answer look like? | “…three short steps, then one practice question…” |
| Constraints | What are the limits? | “…use simple words, no formulas until I ask, and tell me if you are unsure.” |
You do not need all five parts every time. A quick question like “What does CAGR stand for?” needs none of them. Use the full checklist for anything you will study from, send or submit: an email, an assignment brief, a study plan, interview practice.
This is not only our view. Anthropic’s guide also stresses clear, detailed instructions, examples and a stated format.
How do you write each of the five parts?
Write each part as one or two plain sentences that remove a guess the AI would otherwise make.
Role: tell it whose point of view to use
A role tells the AI which kind of expert to sound like. “Explain GST as a chartered accountant advising a small shop owner” gets a more focused answer than “Explain GST”.
A role changes the point of view, not the facts. If you tell the AI it is a Supreme Court lawyer, its case names do not become real. Always check facts yourself.
Task: one verb, one job
Give each prompt one main job. A common mistake is asking for four things at once: “summarise this chapter, make flashcards, explain the hard parts and set practice questions.” The answer comes out muddled. Split it into four prompts, each with one clear verb: summarise, explain, compare, critique or rewrite.
Context: the part that helps most
In our view, context improves a prompt more than any other part. The AI knows nothing about you or your course unless you say so. Add who you are, what you already know, what happened and what you want.
Tone: polite and honest, not dramatic.
Keep it under 120 words.
Context means useful background, not private data. Do not paste your Aadhaar, PAN, phone number, address, passwords or marks card into an AI tool. Do not paste other people’s private messages either. Change details where you can: write “my professor” instead of the name.
Format: decide the shape of the answer
If you do not say what format you want, the AI chooses one, usually long paragraphs or too many bullets. Tell it: a table with the columns Term, Meaning and Example; a numbered checklist; a 100-word summary; an email under 120 words. Format instructions are the cheapest way to improve an answer.
Constraints: set the limits
Constraints are the rules the answer must follow: length, tone, what to leave out (“no motivational filler”), what to flag (“tell me any assumption you make”) and permission to say “I am not sure”. That last one matters because an AI can sound confident and still be wrong. This is called hallucination.
A worked example: preparing for an internship interview
A complete prompt names all five parts and gives the AI every fact it needs. This scenario is made up for teaching: a final-year student preparing for a data analyst internship interview.
Task: Run a mock interview with me. Ask one question at a time and wait for my answer.
Context: I am a final-year B.Com student. I have finished a Python and SQL course. My project was a sales dashboard in Excel and Power BI. I have never been interviewed for a technical role.
Format: After each answer, give feedback in three short lines: what was good, what was missing, and a better version of my answer.
Constraints: Start with easy questions and make them harder. Ask five questions in total, then give me a short summary. Use simple English. If you are not sure something is technically correct, say so.
Each part removes a guess. The AI can still make technical mistakes, so check its feedback against your course notes.
How do you improve a prompt that did not work?
Treat prompting as a conversation, not a slot machine: the first answer is a draft that you guide.
The first answer almost never matters. Many beginners treat AI like a slot machine: pull the lever, judge the result, walk away disappointed. A better way is to treat it like a study session with a patient senior: they try, you give feedback, it improves.
Three follow-up moves cover most situations:
- Redirect. “Good explanation, but too long. Cut it to five lines and use an example from cricket.”
- Show, don’t tell. Paste a sample of the style you want, such as your own notes or an answer you liked. AI copies examples better than it follows adjectives like “simple”. This is called few-shot prompting; the research is in Language Models are Few-Shot Learners (Brown et al., 2020), and our advanced prompting guide goes deeper.
- Ask the AI to interview you. “Before you answer, ask me five questions that would help you do this better.” It often asks about details you forgot to mention.
Two or three follow-ups is normal and healthy. If you need ten, your first prompt was missing something. Go back and improve the Context part instead of patching again and again.
Which prompt templates can you use as a student?
Five templates cover much of a student’s routine writing and studying. Copy them and replace the brackets with your own details.
Context: [who I am, what happened, what I am asking for].
Tone: [polite and formal]. Under 120 words. Do not apologise more than once. Do not invent facts I have not given you.
Notes: [paste]
AI can make mistakes in regional languages and in technical details. Check the explanation against your textbook, or ask your faculty.
When should you not use the five-part framework?
Skip the full framework when the task is small, you can check the answer at a glance, or the real problem is not the prompt.
| If your situation is… | The full framework is… | Do this instead |
|---|---|---|
| A quick lookup, such as “What does CAGR stand for?” | ✗ Overkill | Ask it plainly. Check formulas, dates and rules in your textbook anyway |
| You want ideas, such as project topics | ✗ Too early | Start with a loose prompt, see what comes back, then add structure |
| The prompt takes longer to write than the task | ✗ Not worth the time | Just do the task yourself |
| You need facts you cannot check, such as references for an assignment | ✗ Not the fix | A better prompt does not make facts true. AI can invent books and papers, so confirm every source exists |
| You plan to submit AI text as your own work | ✗ Check the rules first | Follow your college’s AI policy. Use AI to learn, then write in your own words |
| An email, study plan or interview practice that really matters | ✓ Worth it | Use all five parts and check the facts before you use the answer |
What do people get wrong about prompt engineering?
Three beliefs cause most of the wasted effort.
“There are secret magic words.”
No. A pack of ready-made prompts cannot know who you are or what you need. A clear structure beats secret words, every time. Save your money.
“Prompt engineering will be obsolete soon, because AI is getting smarter.”
Partly true, mostly not. Newer models guess what you mean better, even from a sloppy prompt. But no AI can read your mind about your course, your level and your goal. Clear briefing is the part that lasts.
“Longer prompts are better.”
Longer prompts that are relevant are better. Extra padding hurts, because everything you write competes for the model’s attention inside its context window, the amount of text it can look at once. Be complete, then stop.
How do you practise this? The prompt rebuild challenge
Rebuild three of your own prompts, then keep the best one. It takes about 25 minutes and uses your real work, which is the point.
- Step 1 (5 min). Find three prompts you typed into ChatGPT, Claude or Gemini in the past month that gave poor answers. Your chat history has them. Be honest; most are short prompts like “explain machine learning”.
- Step 2 (10 min). Rewrite each one using the five parts. Write them out fully (Role, Task, Context, Format, Constraints), even if it feels slow at first.
- Step 3 (5 min). Run the old and new versions in fresh chats. Put the answers side by side.
- Step 4 (5 min). Score both on relevance, specificity, usable-as-it-is and tone, the rubric from our generative AI starter guide. Then note which single part made the biggest difference.
- Keep the best one. Save your best new prompt in a note or doc so you can reuse it. That is the start of a prompt library. Share it with your study group.
- Read next. Advanced prompting: few-shot examples, chain-of-thought and structured outputs covers step-by-step reasoning and answers you can drop into a spreadsheet.
Frequently asked questions
What is prompt engineering in simple terms?
Prompt engineering is writing clear instructions so an AI tool gives you useful answers. It is closer to explaining a task to a classmate than to programming.
What is the best prompt structure for beginners?
Role, Task, Context, Format, Constraints. Say who the AI should act as, the one job to do, the background it needs, how the answer should look and the limits it must follow.
Do I need coding skills to learn prompt engineering?
No. Basic prompting is a plain-language skill, like writing a clear email. You need technical knowledge only when you build applications on top of AI models, for example with embeddings and retrieval.
Is prompt engineering a career in India?
Treat it as a skill rather than a job title. Standalone “prompt engineer” roles exist, but we do not have reliable figures on how many. In our view, employers increasingly expect prompting inside existing roles such as analyst or developer, much like Excel.
Can I use AI for my assignments?
Check your college’s rules first, because they differ. Using AI to understand a topic, practise questions or get feedback is usually different from submitting AI-written text as your own. If you are unsure, ask your faculty.
Is it safe to share personal details with ChatGPT?
Avoid it. Do not paste your Aadhaar, PAN, phone number, address, passwords or marks card, or other people’s private messages. If you must describe a situation, change the details first.
Sources
- Brown, T. et al. “Language Models are Few-Shot Learners.” 2020. https://arxiv.org/abs/2005.14165
- Anthropic. “Prompt engineering overview.” https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview