Affective and Goal-Oriented Factors of Relationship Formation in the Digital Therapeutic Alliance: A Longitudinal Study of Mental Health Chatbots

Paper Title

Affective and Goal-Oriented Factors of Relationship Formation in the Digital Therapeutic Alliance: A Longitudinal Study of Mental Health Chatbots

Publication Info

• Topic area: Digital Therapeutic Alliance in mental health chatbots
• Keywords: Digital Therapeutic Alliance, mental health chatbots, emotional support, practical support, conversational control, trust, satisfaction, Cognitive Behavioral Therapy, user–chatbot relationships

Background and Problem

• Problem / challenge: Existing research on the Digital Therapeutic Alliance (DTA) lacks focused quantitative evidence specific to mental health chatbots. Prior instruments adapted from traditional therapeutic alliance models fail to capture chatbot-specific dynamics such as conversational control and proactive scaffolding.
• Significance: Understanding the relational factors that drive user–chatbot bonding is essential for improving the design and efficacy of mental health chatbots, which are increasingly used as scalable psychological interventions.
• Motivation and related work: While prior studies have explored relational qualities like empathy and trust in chatbots, they have relied on qualitative methods or adapted measures designed for human therapists. This leaves gaps in understanding how chatbot-specific features influence relationship formation and how these relationships impact user well-being.

Solution

• Proposed approach: Development and application of a multi-dimensional survey instrument to model relational and functional dynamics in user–chatbot interactions, tested in a four-week within-subjects study with two CBT-based chatbots, Wysa and Youper.
• Novelty:
  1. Operationalization of relational and functional constructs specific to chatbot interactions.
  2. Identification of two primary drivers of relationship formation: emotional support and practical support.
  3. Clarification that trust and satisfaction are emergent outcomes rather than independent predictors of relationship quality.
  4. Preliminary evidence linking stronger chatbot relationships to improved well-being over time.
• Procedure and key techniques:
  • Selection of two chatbots (Wysa and Youper) with distinct conversational designs.
  • Recruitment of 56 participants for a counterbalanced within-subjects study over four weeks.
  • Development of a survey instrument capturing seven constructs (e.g., emotional support, practical support, conversational control).
  • Psychometric validation of the instrument through exploratory factor analysis and reliability testing.
  • Regression modeling to identify predictors of user–chatbot relationship formation.
  • Exploratory analysis of the association between relationship strength and changes in well-being (WHO-5 scores).

Results

• Concrete findings:
  • Emotional support (β = 0.63, p < 0.001) and practical support (β = 0.32, p = 0.004) were the strongest predictors of relationship formation.
  • Trust (privacy and non-judgmentalness) and satisfaction were highly rated but functioned as outcomes rather than drivers of relationships.
  • Stronger relationships were associated with greater improvements in WHO-5 well-being scores during the later phase of the study (Week 2–4; t = 3.06, p = 0.004).
• Advantage over baselines: The study provides the first quantitative account of DTA structure specific to chatbots, moving beyond therapist-derived models and descriptive studies.
• Experiments / evaluation:
  • Participants alternated between using Wysa and Youper for two weeks each.
  • Survey data were collected at baseline, mid-study, and post-study to assess relational constructs and well-being.
  • Psychometric validation confirmed the reliability and validity of the seven retained constructs.
• Limitations and future work:
  • Limited stylistic and functional variance between the two chatbots constrained the measurement of certain constructs (e.g., input preference, colloquial style).
  • The four-week study duration may not capture long-term dynamics of alliance formation.
  • The sample was not demographically or clinically diverse, limiting generalizability.
  • Future work should explore longer-term studies, richer chatbot designs, and more diverse populations.

Summary

This study quantitatively modeled the Digital Therapeutic Alliance (DTA) in mental health chatbots, identifying emotional support and practical support as the primary drivers of user–chatbot relationship formation. Trust and satisfaction emerged as outcomes of supportive interactions rather than independent predictors. A four-week within-subjects study with 56 participants using Wysa and Youper demonstrated that stronger relationships were associated with improved well-being during later phases. These findings refine theoretical models of the DTA and provide actionable insights for designing chatbots that balance empathy and efficacy. Future research should extend this work to longer-term studies and diverse populations.