Voice delivery enhances the therapeutic relationship between patients and AI therapy agents — Computational Psychiatry 2025

Background

The therapeutic alliance, the quality of the relationship between a person and their therapist, is one of the strongest predictors of outcomes in talking therapy, accounting for roughly 30% of the variance. As AI begins to close the gap between rising demand for mental health care and a limited clinical workforce, a fair question follows: can an AI agent build that relationship, and does the way it communicates, by text or by voice, change the answer?

Methods

In a within-subjects study, 30 UK adults each held two short therapeutic conversations with an AI agent, one by text and one by voice, talking through personal concerns in their own words. Both conditions used identical prompts; only the modality differed. The team measured therapeutic alliance with the Working Alliance Inventory (WAI-SR), alongside emotional support, engagement, and each person's modality preference.

Results

Voice produced a significantly stronger therapeutic alliance than text (d = 0.56, p = 0.005), and two-thirds of participants reported a stronger relationship with the voice agent. The effect was larger for emotional support (d = 0.62, p = 0.002), preferred by 70%. People also opened up more: spoken turns were about twice as long as typed ones, a sign of deeper disclosure. The gains landed where they matter most for therapy, in relationship and emotional support, rather than in general usability, which showed no significant difference.

Conclusions

For talking therapy, the relationship drives outcomes, and voice strengthens it. That is why Limbic Care delivers therapy by voice: to build a strong therapeutic relationship with every person from the first conversation.