How to write AI prompts: a simple beginner framework
Learn how to write AI prompts with a simple framework for clearer answers, useful examples for work and study, and practical ways to check each result.
DIT Team 6 min read
To learn how to write AI prompts, tell the assistant what you need, who the answer is for, what context matters and what form the result should take. A clear prompt gives the tool a useful starting point, but you may still need to ask follow-up questions and check the answer. This beginner framework works for everyday tasks at work, college or home.
If you are new to AI, join the free AI 101 class. It covers ChatGPT, Claude and Gemini on real tasks, teaches a reusable prompt framework and shows where answers need checking, with no coding needed.
What is an AI prompt?
A prompt is the instruction or question you type into an AI assistant. It can be a short request, such as asking for a recipe idea, or a detailed description of a task, audience and format. The more useful context you provide, the less the assistant has to guess.
A prompt does not guarantee a correct answer. AI tools can misunderstand your request, miss important context or produce a confident but incorrect detail. Think of the result as a first draft that you read, improve and verify before relying on it.
How to write AI prompts with a simple framework
Use these four parts when you want a more useful answer:
- Task: say what you want the assistant to do.
- Audience: explain who will read or use the answer.
- Context: include relevant facts, limits or background.
- Format: tell it how to organise the response, such as a short message, checklist or table.
You can add a fifth part, quality checks, when accuracy matters. Ask the assistant to flag missing information or list facts you should verify. Do not ask it to invent anything that you have not supplied.
A vague prompt and a clearer prompt
A vague request might be: "Write a post about my shop." It does not say what the shop sells, who the post is for or what the reader should do next. The assistant may fill those gaps with assumptions.
A clearer version might be: "Draft a short Instagram caption for a family-run bakery in Pune. The audience is local customers who enjoy fresh bread. Mention only these confirmed details: we sell sourdough and whole-wheat loaves, and orders are collected in the shop. Keep the tone warm, use simple English, and do not add prices or opening hours. After the draft, list the details I should verify."
The second prompt gives the tool a task, audience, context, format and limits. You still need to review the wording and confirm the details, but you have reduced the amount of guessing.
Try this now
Choose a small task from your day, such as rewriting a work message or planning a revision session. Copy this prompt and replace the square brackets:
Help me [task] for [audience]. Here is the context you may use: [relevant facts]. Keep the answer [length or tone] and format it as [message, list or steps]. Do not add facts I have not provided. If you need more information, ask me before making assumptions. At the end, tell me what I should check.
After you get an answer, check it against the task. If it is too long, ask for a shorter version. If it uses the wrong tone, describe the tone you want, such as polite, direct or friendly. Change one instruction at a time so you can see what improved the result.
Improve the answer with follow-up prompts
You do not need to write a perfect prompt on the first try. Treat the first reply as a draft, then give a focused follow-up: "Make the second paragraph shorter," "Explain the second step with an example," or "Use simpler words for a beginner." Specific changes are easier to judge than a broad request such as "Make it better."
If the assistant gives an answer that is too confident, ask what assumptions it made and which details need checking. When the task depends on current prices, schedules, policies or rules, verify those against the relevant official source. Do not assume that repeating the same question will make an answer true.
Common prompt-writing mistakes
- Leaving out the audience: an answer for a customer may need a different tone from notes for your own use.
- Combining many jobs: asking for research, a long report, a social post and a translation at once can make the output hard to review. Split a large task into steps.
- Giving sensitive details: do not paste passwords, bank details, personal ID numbers or confidential customer information into a chat.
- Asking for facts without a source: the assistant may guess. Provide trusted material when appropriate, and ask it to stick to that material.
- Accepting the first answer: read it carefully, edit it and check important claims before using it.
Use the same framework in different situations
For a student, a useful prompt could ask for a concept to be explained in simple steps, followed by practice questions. The student can answer first, then compare the explanation with their notes or textbook.
For a small business owner, a prompt could request a draft reply to a common customer question using only confirmed product and delivery information. The owner should remove private data and review the response before sending it.
For a job seeker, a prompt might help organise experience into a clear outline for a CV. The person must make sure every skill, date and responsibility is accurate and keep the final wording truthful.
The same framework applies across these examples: task, audience, context, format and checks. You can try it with ChatGPT, Claude or Gemini, then compare which result is easiest to understand and verify. Features and limits can change, so check the current options in the tool you use.
For a wider beginner plan, read how to learn AI for free. If you want guided practice with Claude prompts, the Master Claude workshop costs ₹99 + GST and runs live on Zoom on Mondays from 7 to 10 pm IST.
Frequently asked questions
What should a good AI prompt include?
A useful prompt says what the task is, who the answer is for, what context matters and what format you want. Add limits or checks when the answer must stick to confirmed information.
Do longer prompts always give better answers?
No. Extra detail helps only when it is relevant to the task. A short prompt with a clear goal can work well, while a long prompt with unrelated instructions can make the request harder to follow.
Can I use the same prompt in ChatGPT, Claude and Gemini?
You can try the same instructions in all three and compare the replies. Each tool may respond differently, and you should still check important details rather than treating agreement as proof.
Keep learning with a free class
Join the free AI 101 class to practise a reusable prompt framework with ChatGPT, Claude and Gemini, with no coding needed. Bring one real task and learn how to refine the prompt and check the answer.
