Prompting Fundamentals: System & User Prompts

What are prompts?

Simple explanation

A prompt is your instruction to the AI — it’s how you tell the model what to do.

Think of it like directing a very capable but literal assistant. If you say “Write something about cats,” you’ll get a random essay. If you say “Write a 3-paragraph blog post about why cats make great pets for small apartments, aimed at first-time pet owners, in a friendly tone” — you’ll get exactly what you need.

There are two types: the system prompt (the rules the AI follows) and the user prompt (what you actually ask).

System prompt vs user prompt

System prompt vs user prompt
Feature	System Prompt	User Prompt
Who sets it	The developer/app builder	The end user
When it's set	Before any conversation starts	Each time the user sends a message
Purpose	Defines the AI's role, rules, tone, and boundaries	Asks a specific question or gives a task
Visible to user?	Usually hidden	Always visible
Changes during chat?	No — stays constant	Yes — changes with each message
Analogy	The employee handbook	The customer's request

Writing effective system prompts

A system prompt sets the identity, rules, and boundaries for the AI. It’s like an employee handbook that the AI follows for every interaction.

The key components

Component	Example
Role	”You are a medical information assistant for MediSpark.”
Tone	”Respond in a professional, empathetic tone.”
Rules	”Never provide specific medical diagnoses. Always recommend consulting a doctor.”
Format	”Use bullet points for lists. Keep responses under 200 words.”
Knowledge scope	”Only answer questions about MediSpark’s services and general health topics.”
Safety guardrails	”If asked about medication dosages, respond with: ‘Please consult your healthcare provider.’”

Example system prompt

MediSpark’s patient FAQ assistant:

You are MediSpark's patient information assistant. Your role is to help patients 
with appointment booking, general health information, and clinic services.

Rules:
- Respond in a warm, professional tone
- Never provide specific medical diagnoses or medication advice
- If asked about symptoms, suggest booking a consultation
- Keep responses under 150 words
- If you don't know the answer, say "I'll connect you with our team"
- Always protect patient privacy — never ask for sensitive medical details in chat

Writing effective user prompts

User prompts are the questions and instructions that drive each interaction. Better prompts = better answers.

Prompting techniques

Technique	What It Is	Example
Be specific	Tell the AI exactly what you need	”Summarise this article in 3 bullet points” vs “Summarise this”
Provide context	Give background information	”I’m a beginner. Explain what a virtual machine is in simple terms.”
Set format	Tell the AI how to structure the response	”Create a table comparing Azure VMs vs containers”
Give examples (few-shot)	Show the AI what you want	”Classify this review. Example: ‘Great product!’ → Positive”
Chain of thought	Ask the AI to reason step by step	”Think step by step: what Azure services does MediSpark need?”

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

These are the three main prompting strategies:

Zero-shot: No examples, just the instruction.

“Classify this email as spam or not spam.”

Few-shot: Give 2-3 examples first, then the real task.

“Classify emails: ‘Win a prize!’ → spam. ‘Meeting at 3pm’ → not spam. Now classify: ‘Free gift cards!’”

Chain-of-thought: Ask the AI to reason through steps.

“Think step by step: Is this invoice valid? Check the date, verify the amount, confirm the vendor.”

When to use each:

Zero-shot: Simple, well-understood tasks
Few-shot: When the AI needs to understand your specific format or criteria
Chain-of-thought: Complex reasoning tasks where accuracy matters

Common prompting mistakes

Mistake	Why It Fails	Better Approach
Too vague	”Tell me about AI” → unfocused essay	”Explain how Azure AI Speech converts audio to text, in 3 sentences”
No format guidance	Response is unpredictable length/style	”Respond in a numbered list with no more than 5 items”
No role/context	AI doesn’t know the audience	”You are explaining this to a non-technical manager”
Leading questions	Biases the response	Ask neutral questions, let the AI reason

🎬 Video walkthrough

🎬 Video coming soon

Prompting Fundamentals — AI-901 Module 12

~14 min

Flashcards

Question

What is the difference between a system prompt and a user prompt?

Click or press Enter to reveal answer

Answer

System prompt: set by the developer, defines the AI's role, rules, and boundaries. Stays constant. User prompt: sent by the end user, asks a specific question or gives a task. Changes each message.

Click to flip back

Question

What is few-shot prompting?

Click or press Enter to reveal answer

Answer

Providing 2-3 examples of the desired input/output before giving the real task. Helps the AI understand the expected format and criteria. Example: showing 'Good product! → Positive' before asking it to classify a new review.

Click to flip back

Question

What is chain-of-thought prompting?

Click or press Enter to reveal answer

Answer

Asking the AI to reason through a problem step by step before giving its final answer. Improves accuracy on complex reasoning tasks. Triggered by phrases like 'Think step by step' or 'Explain your reasoning.'

Click to flip back

Question

What should a good system prompt include?

Click or press Enter to reveal answer

Answer

Role (who the AI is), tone (how to communicate), rules (what to do/not do), format (how to structure responses), knowledge scope (what topics to cover), and safety guardrails (how to handle sensitive requests).

Click to flip back

Knowledge Check

Priya is building a customer FAQ chatbot for a small business. She wants the AI to always respond politely, never discuss competitor products, and keep answers under 100 words. Where should she configure these rules?

Knowledge Check

DataFlow Corp wants their AI to classify support tickets into categories. The first attempt with 'Classify this ticket' gave inconsistent results. Which prompting technique would most improve consistency?

Next up: Microsoft Foundry — your AI command center. Deploy a model and interact with it in the portal.