Send the same user question with three different system prompts (a pirate, a maths tutor, a five-year-old) and compare the replies. Discuss how the system prompt sets behaviour without the user seeing it.
Exercise (2)
Write a system prompt for a character of your choice.
Explain that LLMs organize conversations using message roles:
system (hidden rules/persona),
user (human input), and
assistant (AI output).
Guide Exercise 1 by showing how changing the system prompt alters tone, vocabulary, and response style while keeping the user query identical.
Encourage creativity in Exercise 2, helping students write specific system boundaries (e.g., "Answer only using emojis" or "Act like a helpful detective").