Studies find AI therapy replies often score higher on empathy than human clinicians

sapinker · x · 2026-07-29

A long Substack post reviews research on whether AI chatbots or human therapists give better therapy-style responses.

The article highlights a 15-study review on empathy: in 13 studies, AI responses were rated as significantly more empathic, with pooled results suggesting ChatGPT-3.5 and GPT-4 scored about two points higher on a ten-point empathy scale. It also notes an important limitation: most comparisons are text-only and cannot capture eye contact, tone, or body language. The piece then turns to a preprint on Mind Companion, an AI psychotherapy support system that analyzes emotions, facts, and risks before generating responses, using it as another data point in the broader debate about whether bots can outperform humans in some therapy-like interactions.

Original post →

More from AGI Musings

AGI Musings channel →