Export your worksheet
Exports include your worksheet, both generated prompts, and your test notes. Save progress to keep working later; open that file to pick up where you left off.
The framework
Start with what students should be able to do, then build five layers. Each layer answers one design question, and each builds on the one before it.
- 0Learning goalWhat should students be able to do afterward?One observable skill: “Students will [skill] by [talking with whom] in [what situation].”
- 1WorldWhere are they, and what is at stake?Setting, the student’s role, the stakes, and the constraints (time, rules, realism).
- 2CharacterWho are they talking to?Identity, motivation (what they want from this conversation), voice, and boundaries.
- 3KnowledgeWhat does the character know, and when do they share it?Four tiers: shares freely, shares if asked well, guards, doesn’t know. Each fact gets an unlock condition.
- 4ModalityWill students type or talk?Voice for real-time conversational skills, text for careful reasoning and clean transcripts, or both.
- 5Evaluation and feedbackHow will students know how they did?How the character reacts in the moment, a separate AI coach with your rubric, and your own debrief.
Three core principles across the layers
- Give the character a want, not just a role.AI models default to being helpful and agreeable. Motivation creates the pushback that makes practice real.
- Knowledge gating is where the learning happens.If the character shares everything on the first question, there is nothing to practice.
- Separate the actor from the judge.The character stays in role. A different prompt grades the transcript.
Example: The Potluck Outbreak
An interview simulation for an Intro to Epidemiology course. Students play a county health investigator questioning a diner owner after a foodborne illness outbreak.
Your worksheet
Fill in one layer at a time for a single case. The generated prompts update as you type. Your work saves in this browser automatically when the browser allows it; use Export to keep a copy.
Test it (a lot) before students do
Spend ten minutes trying to break your simulation. When a test fails, the layer named on the right is usually where to fix it. Check off each test and note what happened.
Paper prototype first. Hand your worksheet to a colleague and ask them to play the character from it alone. Anything they have to guess or invent is a gap the AI will fill on its own.
Before deploying, check your institution’s guidance on approved AI tools and on student data in transcripts.