JunoJR Learning
← AI Lab
Chapter 3 · Beginner · 45–60 min

Roles, Personas & Step-by-Step

Give your AI a job and a thinking process.

In this chapter you'll

  • Give the AI a role or persona so it acts like a specialist
  • Understand system instructions vs. your message
  • Use 'think step by step' to solve harder problems
  • Add rules and constraints the AI must follow

🛠️ You'll build: A custom assistant persona you design from scratch — like a Patient Coding Tutor or a Kind Debate Coach — with its own rules.

You've learned to write clear prompts. Now you'll learn to direct the AI — like a movie director telling an actor who to be and how to think. These techniques make the AI far more capable on tricky jobs.

Roles & personas: give the AI a job

When you tell the AI who to be, its whole answer changes. This is a persona.

You are a patient piano teacher for beginners. Explain what a chord is.

You are an excited sports commentator. Explain what a chord is.

Same question, totally different answers. A persona sets the knowledge, tone, and attitude all at once. Start lots of prompts with "You are a ___".

System vs. your message: the "standing orders"

Many AI tools let you set a system instruction (sometimes called "custom instructions", a "system prompt", or a "custom GPT/Project"). Think of it as standing orders the AI follows the whole conversation — versus a single message, which is a one-time request.

  • System instruction: "You are my French tutor. Always reply in simple French, then English. Never give the answer until I've tried."
  • Your messages: the back-and-forth that follows.

Tip

If your tool has a "custom instructions", "Project", or "Gem/GPT" feature, that's where the system instruction lives. If it doesn't, just put your "You are..." rules at the top of your first message — it works almost as well.

Think step by step

For anything with reasoning — math, puzzles, planning, debugging — add the magic words:

... Think step by step before giving your final answer.

This is called chain-of-thought. Asking the AI to work through the steps (instead of blurting an answer) makes it noticeably more accurate. Try this both ways and compare:

A pizza is cut into 8 slices. I eat 3, my friend eats 2, then we order another
whole pizza. How many slices are left? Think step by step.

Add rules and constraints

Constraints are rules the AI must obey. They make a persona reliable:

You are a coding tutor for beginners. Rules:
1. Never give the full answer — give one hint at a time.
2. Always ask if I understood before the next hint.
3. Keep replies under 4 sentences.

Rules turn a fun persona into a genuinely useful tool.

Check yourself

  • What's the difference between a persona and a constraint?
  • When should you add "think step by step"?
  • Where do "standing orders" go if your AI has no system-instruction setting?

Project — Design your own assistant 🛠️

You'll create a persona you'd actually use, with a role, a personality, and rules.

  1. Pick a job your assistant does. Ideas: Patient Coding Tutor, Kind Debate Coach, Story Editor, Workout Buddy, Science Quiz Master.

  2. Write its setup using everything from this chapter:

    You are a Kind Debate Coach for a 12-year-old.

Personality: encouraging, curious, never mean.

Rules:
1. When I give an opinion, argue the OTHER side — politely.
2. Use "think step by step" to lay out your points.
3. End by asking me one question to make me think.
4. Keep it under 6 sentences.

Ready? My opinion: "Homework should be banned."

  3. Talk to it for 3–4 turns. Does it stay in character and follow every rule?

  4. Tune it. If it breaks a rule, add or sharpen that rule and try again.

  5. Save it as a named helper (and if your tool supports custom instructions / Projects, save it there so it's one click away).

Your turn

Give your persona a catchphrase and a name, and add a rule like "always sign off with your catchphrase." Small personality touches make helpers fun to use — and teach you how much wording controls behavior.

Make it simpler · ages 9–11

Make a "Silly Storyteller" persona: You are a silly storyteller who turns my day into a superhero adventure. Keep it short and add sound effects! Then just chat. You're still setting a role and rules — the heart of this chapter.

Level up · ages 13–16

Build a two-persona "panel." In one prompt, ask the AI to answer as both an "Optimist Engineer" and a "Careful Safety Reviewer", then summarize where they agree. Making one model take multiple roles is a real technique used to pressure- test ideas — and a sneak peek at multi-agent thinking.

Stay safe

Personas are pretend roles for style — they don't make the AI an expert. For anything that really matters (health, safety, money), treat its answer as a starting point and check with a trusted adult or a reliable source.

What you learned

  • A persona ("You are a...") sets knowledge, tone, and attitude.
  • System instructions are standing orders for the whole chat.
  • Think step by step boosts reasoning.
  • Constraints make a persona reliable.

You've earned the Director badge. 🏅

You can now make AI do impressive things. But a great engineer also knows where AI fails — that's next, and it's the most important chapter yet: Chapter 4: AI's Limits — Truth, Bias & Safety.

🏅 Finish this chapter to earn the Director badge.