Free Lesson

Synthetic Users You Can Trust: Ground, Test, Validate

Part of The AI Research Capability Stack

30 min
Jul 15, 2026 12:00 PM
Virtual (Zoom)

In this video

What you'll learn

Ground

Build synthetic users from real participant data — interviews, verbatims, motivations — not from a job title and a wish.

Test

Know the boundary. Synthetic users are decent at surfacing objections, weak assumptions, and confusing language. They are unreliable on purchase intent, behavior change, and anything that requires lived context they don't have.

Validate

Run synthetic against real, and treat every divergence as calibration for the next round.

Why this topic matters

Synthetic users can sharpen a study before you recruit a single participant. They can also hand you confident answers no real customer would give. The difference isn't the tool — it's whether you have a way to check. In this Lightning Lesson, John shows the validation skill he teaches inside both of his courses: ground synthetic users in real participant data, learn which questions they answer reliably and which they get wrong, and compare synthetic output against real customers so each divergence improves the next round. You'll leave knowing when to trust a synthetic answer, when to override one, and how to make that call visible to your team.

You'll learn from

John Whalen, PhD

John Whalen, PhD

CEO, AI Research Leader, Cognitive Scientist, O'Reilly Author

Key Clients
Google
Capital One
Bloomberg
GSK
Pinterest
See all products from John