Abstract
BACKGROUND: Parents of autistic children frequently experience elevated stress levels, depressive symptoms, and reduced well-being. Positive psychology interventions (PPIs) can strengthen resilience, and chatbots offer a scalable channel through which such skills can be delivered. However, evidence on the evaluation of large language model-guided PPI-based chatbots for this population is limited. OBJECTIVE: This study evaluated the feasibility and acceptability of a GPT-powered chatbot ("Allie"). This study was designed to deliver culturally adapted PPIs to parents of autistic children and to explore their preliminary effects on well-being, depression, stress, and health-related quality of life. METHODS: We conducted a single-arm mixed-methods pilot study with 19 parents with autistic children. These parents engaged with Allie for 2 weeks to complete 8 structured PPI exercises. The primary outcomes were feasibility (completion, ease of use, and practicality) and acceptability (multidimensional user ratings). Secondary outcomes were the World Health Organization-Five Well-Being Index (WHO-5), Patient Health Questionnaire-9, Perceived Stress Scale-10, and Short Form-12 Health Survey (version 2) Physical and Mental Component Summary scores. Outcomes were analyzed using paired t tests or Wilcoxon signed-rank tests. Optional postintervention interviews were analyzed using reflexive thematic analysis. RESULTS: A total of 17 (89.5%) participants completed all the exercises, which indicated a high degree of procedural feasibility. There were also high ratings for ease of use and practicality (means 4.47/5, SD 0.70, and 4.32/5, SD 0.67, respectively). Acceptability was favorable (overall satisfaction mean=5.68/7, SD 0.70; prompt response time=6.37/7, SD 0.68). The WHO-5 score improved significantly from 32.84 to 46.11 (t(18)=2.48, P=.02; Cohen d=0.52). Changes in the Patient Health Questionnaire-9 (z=-0.49, P=.63; r=0.11), Perceived Stress Scale-10 (t(18)=-0.82, P=.43; Cohen d=0.12), and Short Form-12 Health Survey (version 2) Physical Component Summary (t(18)=-0.94, P=.36; Cohen d=0.18) and Mental Component Summary (t(18)=-0.89, P=.39; Cohen d=0.17) scores were not significant. Qualitative feedback (14/19) described benefits aligned with PPI mechanisms such as greater self-reflection, a more positive orientation, perspective-taking, emotional support, and coping skills. However, participants also suggested refinements, such as more natural conversation (colloquial Cantonese), shorter or less repetitive outputs, user-chosen sequencing with reminders and progress tracking, multimodal features, and autism spectrum disorder-specific resources. CONCLUSIONS: This pilot study revealed the feasibility, acceptability, and preliminary improvement in well-being (WHO-5) of a PPI-based GPT-powered chatbot, Allie, among parents of autistic children. Although there was no significant short-term change in other outcomes, the findings provide insights into design priorities, including personalization, conversational naturalness, multimodal content, and autism spectrum disorder-specific guidance. Larger, controlled trials with longer exposures and more diverse samples are needed to establish efficacy and durability. TRIAL REGISTRATION: ClinicalTrials.gov NCT06438120; https://clinicaltrials.gov/study/NCT06438120.