1282 Assessing the Effect of an AI-Based Chatbot Intervention on Sleep

Abstract Introduction Poor sleep is a significant public health concern in the U.S. Previous sleep interventions often face barriers such as high costs, limited accessibility, and low user engagement. To address these gaps, our team developed an artificial intelligence (AI)-driven chatbot powered by large language models that delivers individualized guidance on sleep promotion. This study examined the preliminary efficacy of the AI chatbot on sleep outcomes. Methods We conducted a two-group parallel randomized controlled trial among U.S. adults with short sleep and poor sleep quality. Participants were randomly assigned to either the intervention group (two-week engagement with a virtual sleep therapist via text messaging) or the waitlist control group (two-week wait followed by the intervention). Sleep outcomes included Fitbit-measured total sleep time (TST) and self-reported sleep quality, insomnia severity, sleep environment, and sleep hygiene, assessed using validated instruments. Results This ongoing trial has enrolled 12 adults (4 control and 8 intervention) to date. The participants’ mean age was 42 ± 10 years, and 42% were female. At baseline, all participants reported poor sleep quality, with an average of TST of 319 ± 47 minutes. No significant between-group differences were observed across any sleep measurements. After engaging with the AI chatbot, the intervention group showed a significant increase in TST (intervention vs. control: + 60.3 vs. -19.5 minutes, p = 0.004) and a greater reduction in poor sleep hygiene behaviors (intervention vs. control: -26.8 vs. 1.3) than the control group. Both groups demonstrated significant improvements in TST (mean difference MD: 58.0 ± 39.6 minutes, p 0.001), perceived sleep quality (MD: -4.6 ± 4.0, p = 0.002), insomnia severity (MD: -7.7 ± 6.4, p = 0.002), and sleep hygiene behaviors (MD: -22.3 ± 8.8, p 0.001) after the intervention. Conclusion The preliminary findings suggest that our AI-based chatbot may have a positive effect on sleep promotion. A fully powered sample is needed to confirm these early results. If replicated, this work highlights the promise of integrating generative AI technologies into sleep interventions. Support (if any)