With the widespread application of artificial intelligence agents in daily life, social cues have received increasing attention as an important factor affecting human-machine trust. However, its specific role in depression treatment chatbots still needs further study. This study aimed to investigate how social cues influence users’ trust in a depression intervention chatbot. In the experiment, we compared two groups of chatbots: one therapeutic chatbot with high levels of social cues, using avatar animation to express emotional tone, facial expressions, and body movements, and the other chatbot with low levels of social cues. A bot that uses humane and empathetic text conversations. We also collected participants’ skin conductance and heart rate data to further explore the relationship between these physiological indicators and trust. The results of the study showed that in the self-help intervention for depression, therapeutic chatbots with higher levels of social cues significantly improved users’ trust levels and showed clear advantages in forming therapeutic alliance and user satisfaction. Specifically, the high social cue agent group scored significantly higher on the dimensions of perceived reliability (p < 0.05), perceived technical competence (p < 0.01), and belief (p < 0.01) compared to the low social cue agent group higher. In addition, the high social cue agent group had significantly higher heart rate (d = 0.90) with a large effect size, and skin conductance (d = 0.71) had a medium effect size. Significant changes were observed in both heart rate (F = 48.63, p < 0.01) and skin conductance (F = 26.25, p < 0.01). In summary, the findings of this study not only reveal the key role of social cues in depression intervention chatbots, but also provide valuable practical guidance for improving human-computer interaction design in the future.