Digital Warmth but not Digital Hope? How AI- vs. Therapist Guidance shape Common Factors in Digital Mental Health Interventions

Background: Human guidance is one of the most effective components of digital mental health interventions (DMHIs). The emergence of artifical intelligence (AI) capable of generating empathic, contextually relevant responses raises the question of whether AI-generated feedback can replicate the therapeutic mechanisms that make human guidance effective, and whether users would accept it. Methods: In a preregistered 2×2 vignette experiment, N = 116 participants rated feedback responses generated either by a human therapist or an AI on five common factors of psychotherapy (therapeutic alliance, empathy, expectations, congruence, and competence) as well as their acceptance of DMHIs. Feedback source (human vs. AI) and transparency (disclosed vs. undisclosed) were manipulated between participants. Results: Contrary to our hypothesis, AI-generated feedback did not outperform therapist-written feedback on the five common factors. Therapist feedback was perceived as more effective at instilling hope and belief in improvement, regardless of the transparency condition. Disclosure of AI authorship did not significantly affect common factor ratings. Participants in the AI disclosed condition reported lower expected helpfulness of AI-based DMHIs than undisclosed participants. Conclusions: AI-generated feedback can mimic most therapeutic common factors but falls short on expectation enhancement. Transparency about AI involvement does not undermine perceived feedback quality but may reduce openness to DMHIs.