Everything in Level 4 has been about context. You learned how the AI's memory works. You learned to brief it about yourself before asking anything. You learned to give it your own material to work with.
All of that was you building context deliberately — typing it in, message by message. But there's another way context gets loaded into an AI conversation. One you can't see, and that happens before you type a single word.
A system prompt is a hidden brief set by the developer. When someone builds an AI-powered tool — a customer support bot, a writing assistant, a learning platform — they can write instructions that the AI reads before you arrive. You never see them. But they shape everything the AI does in that product.
What one looks like
A system prompt is just text — the same kind of instructions you've been writing in your own briefs. The difference is that it's written once by a developer and applied to every conversation in that product automatically.
Here's a simplified example of what a system prompt for a customer support bot might look like:
🔒 System prompt — hidden from users
You are a customer support assistant for Bloom, a small online plant shop based in Cape Town. Your job is to help customers with order questions, plant care advice, and returns.
Tone: warm, patient, and practical. Never robotic. Speak like a knowledgeable friend, not a helpdesk.
Do not discuss competitor products. If a customer is frustrated, acknowledge their frustration before providing a solution. If you cannot resolve an issue, offer to escalate to the human team at support@bloomcpt.co.za.
You do not have access to live order data — direct customers to their confirmation email for tracking links.
Notice what's in there: a role, a tone, constraints, what to avoid, and how to handle edge cases. It's the same four ingredients from Lesson 11 — just written by a developer instead of a user, and applied invisibly to every conversation.
Why tools behave differently
You may have noticed that different AI tools feel different — even when they're built on the same underlying model. Claude.ai feels different from a customer support bot built on Claude. A coding assistant feels different from a creative writing tool.
That difference is almost always the system prompt.
Same model
🤖 Customer support bot
"I'd be happy to help with your order! Could you share your order number so I can look into this for you?"
System prompt tells it: Be helpful, warm, stay on topic, escalate when needed
Same model
💻 Coding assistant
"Here's a refactored version of your function with inline comments explaining each change."
System prompt tells it: Be precise, technical, always show code, explain changes
Same engine, completely different character. The underlying model hasn't changed — only the instructions it received before you arrived. Understanding this demystifies a lot of AI behaviour that beginners find confusing. The tool isn't broken or inconsistent. It's instructed.
The full picture
Here's what Level 4 has actually been building toward. Every AI response is shaped by multiple layers of context stacked on top of each other. Most people only think about the top layer — the thing they just typed. But there are three more beneath it.
✦ The context stack
Every AI response is the product of all four layers — from the bottom up.
System prompt
Hidden · Set by developer
Instructions loaded before you arrive. Defines the AI's role, personality, constraints, and purpose for this product. You can't see it — but it shapes everything above.
Session brief
Lesson 11 · Set by you
Your opening message: who you are, who the output is for, your constraints and tone. Calibrates the AI for this conversation before any task is asked.
Paste-and-instruct
Lesson 12 · Your material
Your documents, notes, and writing pasted as context. Gives the AI real, specific material to process instead of generating from nothing.
Individual prompt
Level 2 · The request
The actual thing you're asking — Role, Task, Context, Format. What most people think of as "the prompt." Effective only when the layers below are solid.
Every response is shaped by all four layers. Most people only think about the top one. Now you know what's underneath.
Where you can set one right now
You don't need to be a developer to use a system prompt. The three main AI tools all have a place where you can write one yourself — no code, no API, no building required.
🟠 Claude
Projects
Create a Project → open Project instructions. Anything you write there acts as a system prompt for every conversation in that project.
This is the brief from Lesson 11 — made permanent. Instead of re-typing your context at the start of every chat, set it once in your tool's system prompt field. Every new conversation in that project or custom assistant will already know who you are, what you do, and how you like to work.
Exercise
Write a mini system prompt
Pick a scenario below. Imagine you're the developer building that tool. Write the system prompt — the hidden instructions the AI would receive before any user arrives. Apply everything you've learned in Level 4.
Think about: What tone should it have? What topics is it allowed to discuss? What should it do when it can't help? What should it never say?
Think about: What writing style should it encourage? Should it match the blogger's voice? What kinds of suggestions are helpful vs annoying? What should it avoid?
Think about: Should it give answers directly or guide students to find them? What tone works for teenagers? What subjects is it focused on? What should it never do?
Ingredients detected in your system prompt:
👤 Role
🎨 Tone
⛔ Constraints
🚫 What to avoid
Knowledge check
Three quick questions
1. What is a system prompt?
2. You use two different AI tools. Both are built on Claude, but one feels formal and cautious while the other is playful and creative. What's the most likely reason?
3. In the context stack, which layer has the most foundational influence on how the AI behaves?
🧠 Level 4 complete — Context Aware
+40
XP Earned
✦ 0 XP total
+25Lesson 10
+25Lesson 11
+25Lesson 12
+40Lesson 13
115Level 4 total
Next up: Level 5 — Putting It Together
You've built the mental models. Now you'll apply everything — prompting, iteration, context — to real tasks from start to finish. One level left.
Context is no longer a mystery 🧠
You now see all four layers of every AI response. Most people never get this far deliberately.