3/31/2026 · 46 min
Can a Chatbot Really Care? Woebot's Founder on Therapeutic Bonds, CBT, and What LLMs Still Can't Do
Prove It! with Dr. Ali Darcy, Founder, Woebot Health
Audio coming soon · 46 min
Before mental health chatbots were a trend, Dr. Ali Darcy was already running randomised controlled trials on one. In 2016, she and her collaborators at Stanford conducted what is widely regarded as the first ever formal RCT of a conversational AI for mental health — a paper since cited over 2,000 times. Her creation, Woebot, became one of the most studied and recognised digital mental health tools in the world.
In this episode, Paul Wicks sits down with Ali for a wide-ranging, deeply honest conversation about what the science of AI mental health actually shows, and what it doesn't — the engagement crisis in digital CBT, why humour is a legitimate and underused therapeutic tool, the surprising finding that users form a meaningful bond with a fictional robot character, and the question the field still hasn't answered: what exactly is the active ingredient of the human therapist?
Topics covered
- The 2016 Stanford RCT: design choices that still hold up nearly a decade later
- Engagement as the primary unsolved problem in mental health tech
- The therapeutic alliance paper: 37,000 users, 200x larger than the entire existing literature
- What Woebot got right by being transparently fictional rather than pretending to be human
- Why LLMs can generate profound-sounding insights that users forget within 24 hours
- The Goldilocks zone of evidence: how much is enough, and when does more actually hurt you?
- Role confusion as a safety crisis: why being a companion and a therapist is clinically dangerous