Prompt Engineering Fundamentals (The Only Framework You Need)

Prompt engineering isn't a hack — it's the user interface of every LLM. Once you understand the five primitives below, you can write prompts for any model (ChatGPT, Claude, Gemini, Llama) without memorizing tricks.

The 5 primitives of a great prompt

1. Role — who the model should be.
2. Context — what it needs to know.
3. Task — the one outcome.
4. Format — the shape of the answer.
5. Constraints — what to avoid, limits, tone.

Zero-shot vs few-shot vs chain-of-thought

Zero-shot: just ask. Works for easy, common tasks.
Few-shot: include 2–4 examples of input → output. Best for tone matching, classification, structured extraction.
Chain-of-thought: ask the model to reason step by step before answering. Best for math, planning, and multi-step decisions.

The cheat sheet

• "Act as…" sets role.
• "Think step by step" unlocks reasoning.
• "Return as a markdown table with columns X, Y, Z" forces structure.
• "Before answering, ask me clarifying questions" prevents bad answers.
• "Critique your own answer, then improve it" doubles quality on hard tasks.
• "Be concise. No filler." beats most "be brief" phrasings.

A template you can steal

# Role
You are {role with seniority}.

# Context
- Audience: {who}
- Background: {what they need to know}
- Why this matters: {goal}

# Task
{one sentence, one outcome}

# Format
{shape — bullets, table, JSON schema, 200-word email…}

# Constraints
- {tone}
- {what to avoid}
- {length}

# Examples (optional)
Input: …
Output: …

How to debug a bad answer

1. Did you give a role? Add one.
2. Did you give an example? Add one.
3. Did you tell it what not to do? Add the anti-pattern.
4. Is the task multi-step? Ask for a plan first, then execution.
5. Is it hallucinating? Lower the scope ("only use the text I pasted").

Prompt engineering is just clear thinking, written down. If you can brief a junior teammate, you can prompt an LLM. The teams that win are the ones who write their best prompts down, version them, and reuse them — turning prompts into internal software.