Campus mental health services are chronically under-resourced, leaving millions of students without timely support between therapy sessions. A promising model now under evaluation uses conversational AI not as a therapist replacement but as a structured between-session companion — a distinction that matters enormously for both safety and efficacy in clinical design.
This single-arm pilot enrolled university students aged 18 and older presenting with moderate subclinical symptoms — Social Phobia Inventory scores of 21–40, PHQ-9 scores of 5–14, or GAD-7 scores of 5–14 — but excluding those with active psychiatric diagnoses, suicidal ideation, or psychotropic medication use. Participants completed four weekly group sessions following the Unified Protocol (UP), a transdiagnostic CBT framework, while gaining access to a Claude 3.7-Sonnet large language model chatbot programmed with UP-aligned therapeutic prompts. The chatbot served explicitly as a between-session adjunct, not a standalone intervention. Primary outcomes tracked treatment adherence, chatbot engagement, and system usability via the System Usability Scale; secondary outcomes measured symptom changes across anxiety and depression scales.
What distinguishes this design from earlier mental health chatbot studies is the deliberate embedding of the AI within an established clinical protocol rather than testing it in isolation. This matters because prior standalone chatbot trials — including Woebot and Wysa studies — have shown mixed results partly because they lacked the anchoring structure of concurrent professional care. The UP framework also addresses multiple emotional disorders simultaneously, making it particularly suited to a student population where anxiety and depression commonly co-occur. That said, the single-arm design with no active control limits causal inference, and the subclinical inclusion criteria mean findings may not generalize to students with diagnosable conditions. As a feasibility study, its value lies in informing a larger randomized trial rather than confirming efficacy. Practitioners should view this as an incremental but methodologically careful step toward scalable campus mental health support.
