This paper investigates factors influencing initial and continuous user engagement with emotional AI chatbots for mental health, using technostress theory as the theoretical lens. We conceptualize human interaction, emotional competence and perceived pleasantness as challenge techno-stressors, while perceived risk and system quality defects are treated as threat techno-stressors. User appraisals of technostress, including self-efficacy, personal innovativeness and effort expectancy, are also examined.
A two-step research design was adopted, combining user review analysis and semi-structured interviews with a survey tested through hierarchical regression. Both linear and curvilinear relationships were explored, with particular attention to potential U-shaped effects. Moderator analyses of gender, occupation and education were conducted to capture subgroup differences.
Results indicate that human interaction, self-efficacy, personal innovativeness and perceived risk significantly influenced engagement, with U-shaped effects for human interaction and self-efficacy. Emotional competence and pleasantness showed no main effects but were conditionally important (e.g. students, highly educated users). Gender moderated personal innovativeness, occupation moderated pleasantness, effort expectancy and risk, while education moderated emotional competence, system quality and effort expectancy.
We distinguish initial versus continuous engagement with emotional AI chatbots, extend technostress theory into an emotional health support context and reveal both curvilinear effects and subgroup-specific variations. These findings enrich theory by clarifying the dual roles of challenge and threat stressors and offer actionable guidance for tailoring chatbot design to diverse user groups.
