Stop 12 · Prompt EngineeringTopics · 1/10

Prompt Design Fundamentals

In short

The same model can give you a vague essay or exactly the answer you need. Most of the difference is how clearly you asked. Good prompts are the cheapest upgrade you'll ever make.

Writing clear, specific instructions that reliably get the output you want from an LLM — often the highest-leverage skill for shipping useful AI features quickly.

The blob hands a sample shirt to a tailor and points at a measuring tape on the counter.

Think of it like this

Instructions for the tailor

Tell a tailor 'make me a shirt' and you'll get a surprise. Hand over a sample shirt, say 'slim fit, full sleeves, two pockets, ready by Friday', and you get what you pictured. An LLM is that tailor: skilled, but it can't read your mind.

Watch it happen

Parts of a clear prompt

Role

Task

Context

Format

Limits

1/5Role: 'You are a placement cell mentor.' It sets tone and expertise.

The big ideas

What you need to know

Be specific

Say who it's for, how long, what tone and what to avoid. Vague prompts get average answers.

Separate instructions from data

Wrap pasted text in clear markers like <resume>...</resume> so the model knows what to work on.

Show the output shape

Describe or sketch the format you want. It's easier to parse and easier to check.

Iterate like code

Change one thing at a time and test on several inputs. A prompt that works once may fail on the next.

See it in code

Walk through the code

text

1You are a placement cell mentor at an Indian engineering college.2 3Task: Review the resume summary below for a fresher Data Analyst role.4 5<summary>6Hardworking student skilled in Python, Excel and teamwork. Looking for opportunities.7</summary>8 9Reply in this format:101. Three short bullet points on what to fix112. A rewritten summary under 60 words12 13Do not invent projects or skills that are not mentioned.

1/4Role and task come first, in plain words.

Role, task, clearly marked data, output format and a limit. Each line removes a guess the model would otherwise make.

Save yourself the pain

Common mistakes

Oops

Writing 'make it better' and hoping.

Do this instead

Say what 'better' means: shorter, more formal, more numbers, fewer buzzwords. The model can only aim at a target you name.

Oops

Testing a prompt on one example and shipping it.

Do this instead

Keep 10 to 20 test inputs, including tricky ones, and rerun them whenever you change the prompt.

Your mission

Before and after

~20 min

Quick check

Why wrap pasted user text in tags like <text>...</text>?

Go deeper

1Practice rewriting vague prompts into specific, well-structured ones and compare outputs.
2Learn the core levers: role, task, context, format, constraints.
3Build a personal library of prompts that work well for recurring tasks.