A 15-day study of 38 university students suggests a low-cost social robot can guide cognitive behavioural therapy exercises at home, with early signs of benefit over a chatbot alone.
This is an exploratory, preliminary study, the authors say. Over 15 days, 38 university students used one of three ways to complete CBT exercises: a socially assistive robot powered by a large language model, a chatbot using the same model, or paper worksheets. Self-reported general psychological distress fell significantly over the study in the robot and worksheet groups, but not in the chatbot group. The robot also produced significant single-session anxiety improvements on more days than the other two conditions combined.
The researchers built a low-cost robot that uses an LLM to hold spoken dialogue and walk each participant through standard CBT exercises at home, and compared outcomes against a text-based chatbot and traditional worksheets over the same period.
The authors are careful to flag the exploratory nature of the work. A mixed-effects model found that the robot and chatbot conditions reduced post-session anxiety more for participants who started with elevated anxiety, but the sample is small and the study ran for only 15 days, so the paper treats these results as suggestive rather than conclusive.
For clinicians and students interested in digital mental health tools, the practical message is modest: a robot-delivered format did not clearly outperform worksheets on distress, but it did appear to support more consistent session-by-session anxiety relief than text-only chatbot delivery. The authors frame this as a case for further, larger trials of combining AI personalisation with social robotics, not as evidence of a ready clinical tool.