☕️ Sunday Sharing: How to Build a Virtual Learner to Stress-Test Your Learning Design with AI
This week I built an AI 'virtual learner'—a realistic persona that reviews training like an actual learner.
I fed it an outline for a 45-minute compliance module. The feedback was brutal but priceless.
Here's the 4-step process I used 👇
🪜 Step 1: Collect Real Learner Signals
Grab any 3–5 of these data points:
- Role info > job descriptions or ads
- Real quotes > surveys, interviews, Slack, feedback forms
- Basic context > devices, time, environment
- Performance clues > common errors, KPIs, typical tasks
The more data the better, but a handful of the right sort of info is enough.
🪜 Step 2: Create the Persona
Prompt AI: "You're an expert in instructional design. Using only the information above, create ONE realistic learner persona (~300 words) to give me reliable feedback on my design decisions and ideas. Include: name, role, a day-in-the-life story, goals, frustrations, and when/how they realistically have time to learn."
Rule: If AI invents details you didn't provide, challenge or delete them.
🪜 Step 3: Add "What This Means for the Design"
Prompt: "Add a section called "What this means for the learning experience." For each friction point, provide a concrete solution with rationale. Example: "Too long → Maximum 5 minutes per module because Marcus learns during his commute".
This is the bridge from story → design constraints & decisions.
🪜 Step 4: Use It as Your "Virtual Learner"
Paste the persona into an AI chat and tell it to: "Act as this learner." Then stress-test anything & everything, e.g.
→ Analysis: "Describe three moments this problem shows up in your week."
→ Design: "Look at these draft objectives. Which would you actually complete? What makes you roll your eyes?"
→ Development: "Read this scenario. What feels real? What feels fake? Where would you stop paying attention?"
→ Implementation: "You see this course invite during a normal day. Do you open it? Rewrite the subject line so you'd actually click."
Try these models:
👉Claude Opus 4.5 / Sonnet 4.5 > Exceptional performance against instruction-following benchmarks.
👉GPT-5.1 / 5.2 > Near-flawless at following complex multi-step prompts.
👉Gemini 3 Pro > Extremely strong reasoning power.
We all say we're learner-centered. But when you're moving fast, it's easy to design for the stakeholder deck instead of the learner on the ground.
If you're experimenting with AI in L&D this year, this is one of the safest, highest-leverage places to start.
Happy experimenting and good luck!
Phil 👋
PS: Want to learn more about how to use AI in your workflow? Apply for a place on my AI & Learning Design Bootcamp (link in comments).