Depression intervention using AI chatbots with social cues: a randomized trial of effectiveness

Depression is a serious problem among college students, and chatbots are a popular intervention tool. Social cues are used in chatbot design, but their effectiveness in depression treatment remains to be verified. This study aimed to compare the effects of chatbots with high-social-cue (HSC) versus low-social-cue (LSC) designs on depressive symptoms.

An open-label randomized controlled trial was conducted over 16 weeks. Eighty-four college students with baseline Patient Health Questionnaire-9 (PHQ-9) scores ≥9 were randomly assigned to either an HSC group (text + voice + animations) or LSC group (text-only). Clinical outcomes, including PHQ-9, Generalized Anxiety Disorder scale (GAD-7), and Positive and Negative Affect Schedule (PANAS) scores, were collected every 4 weeks. Secondary measures included user satisfaction (Client Satisfaction Questionnaire-8, CSQ-8), therapeutic alliance (Working Alliance Inventory-Short Revised, WAI-SR), and self-reported adherence.

Baseline characteristics did not differ significantly between groups. Intention-to-treat analysis revealed that the HSC group achieved greater reductions in PHQ-9 (d = 0.63, P < 0.01) and GAD-7 (d = 0.50, P = 0.003) scores compared to the LSC group. The HSC group also demonstrated higher adherence rates (d = 0.82, P < 0.01), CSQ-8 (P = 0.02), and WAI-SR scores (P < 0.001). LSC group.

Chatbots with high-social-cue designs significantly outperformed text-only versions in alleviating depression and anxiety, while enhancing adherence, satisfaction, and therapeutic alliance.