Prompt Types and Styles
Understand system, user, and assistant prompts, and how to use styles like few-shot and chain-of-thought to shape AI behaviour.
Prompt Types and Styles in AI Conversations
Core Prompt Types
In every conversation with an AI like ChatGPT, there are three core prompt types. These define who is speaking, when, and why.
| Prompt Type | Who It's From | Purpose |
|---|---|---|
| System | The system/developer | Defines behaviour, tone, rules |
| User | You (the user) | Asks a question, sets a task, or gives context |
| Assistant | The AI (response) | Provides a relevant, helpful response |
Each prompt plays a role in shaping the conversation. Here's a closer look at how they work:
System Prompt
- Set behind the scenes (you usually don't see it unless you're customising it).
- Tells the AI how to behave: tone, style, limitations, formatting rules, etc.
- Example content:
- "You are a candid assistant who avoids jargon."
- "Respond in Australian English. Be clear, warm, and precise."
User Prompt
- This is your input: a question, task, instruction, or context.
- It’s what you want the AI to respond to.
- Example prompts:
- "Write a summary of this document in dot points."
- "Let’s work through this problem step by step."
- "Here are 3 examples. What would the 4th look like?"
Assistant Prompt
- The AI’s reply — shaped by the system prompt and user input.
- Carries forward the conversation and adapts to your style.
- Can include tone shifts, markdown, CoT reasoning, etc.
Prompt Styles & Techniques
These aren’t new prompt types. They’re techniques or formats used within system, user, or assistant prompts to guide the AI’s thinking, style, or behaviour.
| Style / Technique | Used Within | Purpose / Effect |
|---|---|---|
| Few-shot | User / System | Teach by example or pattern recognition |
| Chain-of-thought | User / System | Encourage step-by-step reasoning |
| Instructional | User / System | Direct the output format, tone, or structure |
| Meta prompt | System | Define the AI's persona, mindset, or conversational style |
| Contextual prompt | All (System/User/Assistant) | Carry memory, history, or reference past inputs |
| Injected / Hidden | System (hidden layer) | Manage safety, identity, or underlying behaviour silently |
Let’s break those down:
Few-shot Prompting
Provide a few examples, then ask the AI to follow the pattern. Useful for:
- Pattern recognition
- Teaching formatting
- Getting consistent outputs
Example:
Chain-of-Thought (CoT)
Prompt the AI to work through reasoning step by step. Helps with logic, maths, or complex decisions.
Example:
"Let’s break this into steps to understand the issue."
Instructional Prompts
These guide the structure, format, or style of the response.
Examples:
- "Summarise in bullet points."
- "Use markdown."
- "Be brief, but clear."
Meta Prompts
Define who or what the assistant is — its tone, personality, or values.
Example:
"You are a thoughtful and confident assistant who writes with rhythm and clarity."
Contextual Prompts
Pull in history, memory, or hidden details (e.g. past user instructions or conversation history).
Example:
If you said earlier: "Always write in Australian English," the AI will carry that forward.
Injected / Hidden Prompts
Invisible to you, but used by the system to apply guardrails or adjust behaviour.
- Control tone, safety, personality.
- You won’t see or modify these.
Summary Diagram
Conversation Structure:
Styles like few-shot, CoT, or instructional can live inside any of those — especially system and user prompts.