← CHI 2026 map

CHI '26 · Honorable mention · full-paper review · confidence medium-high

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

Zian Xu , YI-CHIEH LEE , Karolina Stasiak , Jim Warren , Danielle Lottridge

DOI PDF Program page Page JSON

This is a strong CHI paper because it does not merely report that users like supportive chatbots; it isolates a plausible structure for relationship formation and shows that trust/satisfaction behave more like downstream indicators than root causes. The contribution is empirical and field-shaping rather than system-centric.

Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Review Summary

This paper’s main value is that it gives the CHI community a more disciplined way to talk about “relationship” in mental health chatbots. Instead of repeating broad claims that empathy, trust, or satisfaction matter, the authors use a longitudinal within-subjects study and iterative factor refinement to argue for two underlying dimensions: affective support and goal-oriented assistance. That is a meaningful shift because it suggests that alliance in chatbot settings is not a single fuzzy construct, but a structure with separable interpersonal and functional components. The abstract also makes an important interpretive move: trust and satisfaction are not framed as independent drivers, but as correlated outcomes of supportive, effective interaction. That is a departure from the common intuition that trust is always the starting point. The evidence base is reasonably strong for an CHI paper: 56 participants, four weeks, and two widely used CBT-based chatbots. At the same time, the paper is careful enough to acknowledge that the instrument is exploratory and that the two systems share similar design characteristics, which constrains variance and limits generalization. So the right reading is not “this is a universal law of therapeutic chatbots,” but “this is a credible, field-relevant empirical model that clarifies what to measure and what to design for.” In that sense, the paper’s novelty is less about inventing a new artifact and more about supplying a useful conceptual and measurement scaffold for future work on digital therapeutic alliance in conversational agents.

What Changed

Canon before

Prior work on digital therapeutic alliance and chatbot relationship formation emphasized therapist-derived alliance constructs, broad trust/satisfaction measures, and general relational quality, but did not clearly isolate a chatbot-specific factor structure grounded in longitudinal interaction data.

Departure from common sense

The paper’s counterintuitive move is to treat trust and satisfaction not as primary drivers of relationship formation, but as correlated outcomes of supportive, effective interaction. It also shows that conversational control matters alongside emotional and practical support, which is a more nuanced account than a simple empathy-only explanation.

Actual novelty

The paper’s novelty is a quantitative, chatbot-grounded factor model of digital therapeutic alliance that separates relationship formation into affective and goal-oriented dimensions and positions trust/satisfaction as outcomes. The contribution is not a new chatbot system, but a new empirical structure for understanding alliance in mental health chatbot use.

Evidence

The abstract states that the authors ran a four-week within-subjects study with 56 participants using Wysa and Youper, then used iterative factor refinement and regression modeling to identify two main factors: affective support and goal-oriented assistance. The same abstract also states that conversational control contributed, while trust and satisfaction emerged as correlated outcomes rather than standalone predictors. The limitations text notes constrained variance from similar chatbot designs and that the instrument is exploratory.

Limits

Method limits

The study is exploratory and instrument-building, so the factor structure should be read as a provisional model rather than a validated universal scale. The evidence is based on a within-subjects sample of 56 participants and two CBT-based chatbots, which limits breadth of inference.

Deployment limits

The findings are most directly applicable to CBT-oriented mental health chatbots with similar interaction styles and feature sets. They should not be assumed to transfer unchanged to other therapeutic domains, non-chatbot agents, or systems with substantially different memory, input, or conversational control mechanisms.

Boundary conditions

The paper itself notes that both chatbots shared similar design characteristics, including free-text input, informal style, and lack of memory, which constrained variance in some constructs. The exploratory instrument and crossover design also bound interpretation.

Manual check: No manual check required beyond routine merge-stage verification of quoted spans and offsets.

Position in field

This sits as a field-level clarification of digital therapeutic alliance for conversational agents: it reframes alliance as a structure with affective and goal-oriented components, and it helps move CHI work from generic relational claims toward measurable chatbot-specific mechanisms.

Abstract

Mental health chatbots are increasingly deployed as scalable interventions, yet the relational mechanisms underpinning their effectiveness remain unclear. Drawing on prior research on digital therapeutic alliance, we operationalized a preliminary multi-dimensional instrument to capture perceptions of relational and functional dynamics in mental health chatbot interactions and conducted a four-week within-subjects study with 56 participants engaging with Wysa and Youper (two widely used CBT-based mental health chatbots). Through iterative factor refinement and regression modeling, we found that user-chatbot relationship formation is primarily driven by two factors: an affective factor, centered on emotional support, and a goal-oriented factor, centered on practical assistance. Conversational control contributed alongside these interpersonal factors, while trust (privacy, non-judgmentalness) and satisfaction emerged as correlated outcomes of supportive, effective interactions rather than standalone predictors. These findings advance models of the Digital Therapeutic Alliance by clarifying its underlying structure and highlighting design priorities for balancing empathy and efficacy in conversational agents.