An AI sleep chatbot demonstrated satisfactory feasibility, usability, and acceptability, with preliminary evidence showing improved sleep outcomes including increased total sleep time by 1.4 hours and decreased sleep onset latency by 30.9 minutes, although causality cannot be established.
Key Findings
Results
The AI sleep chatbot showed satisfactory feasibility, with 82.2% of enrolled adults completing chatbot registration, 73.9% initiating interactions, and 50% completing the 2-week intervention.
107 adults enrolled in the study; 88 (82.2%) completed chatbot registration
65 of 88 registered participants (73.9%) initiated interactions with the chatbot
44 of 88 registered participants (50%) completed the full 2-week intervention
Final analysis included 42 adults (mean age 36, SD 11 years; 28.6% male)
The chatbot was integrated into a commercially available messaging app
Results
Participants engaged with the chatbot for an average of 58 minutes total over the 2-week period, with each chat session lasting approximately 9 minutes.
Average total engagement time was 58 (SD 42) minutes
Each chat session lasted approximately 9 (SD 6) minutes
Participants interacted via texting with a virtual sleep therapist over 2 weeks
The chatbot provided individualized sleep guidance and adapted recommendations based on prior conversations
Results
The chatbot received high usability ratings, with an average System Usability Scale score of 85.2 out of 100, well above the benchmark score of 68.
Average usability score was 85.2 (SD 10.7) out of 100
Score exceeded the established benchmark of 68
Most participants reported favorable experiences with the chatbot
Feasibility, usability, and acceptability were descriptively summarized
Results
The chatbot was rated as highly acceptable, with a satisfaction score of 27.3 out of 32, and all participants perceived the chatbot as effective.
Satisfaction score was 27.3 (SD 4.1) out of a maximum of 32
All participants perceived the chatbot as effective
Effectiveness ratings ranged from 'slightly effective' to 'extremely effective'
Most participants reported favorable experiences with the chatbot overall
Results
Total sleep time increased by 1.4 hours and sleep onset latency decreased by 30.9 minutes following the 2-week chatbot intervention.
Total sleep time increased by 1.4 hours (P<.001)
Sleep onset latency decreased by 30.9 minutes (P<.001)
Sleep was assessed using questionnaires before and after the intervention
The study was quasi-experimental with a single group, so causality cannot be established
Results
Sleep efficiency increased by 7.8% and multiple subjective sleep measures showed statistically significant improvement following chatbot use.
Sleep efficiency increased by 7.8% (P=.007)
Perceived sleep quality improved with a mean difference of -5.4 (P<.001)
Insomnia severity improved with a mean difference of -7.9 (P<.001)
Daytime sleepiness improved with a mean difference of -4.7 (P<.001)
Sleep hygiene skills improved with a mean difference of -13.2 (P<.001)
Results
No significant change was observed in sleep environment scores following the chatbot intervention.
Sleep environment mean difference was -1.1 (P=.16), which was not statistically significant
This was the only sleep-related outcome that did not show significant improvement
All other assessed sleep outcomes showed statistically significant improvements
Discussion
The study was a quasi-experimental, single-group design without a control group, limiting causal conclusions about the chatbot's efficacy.
Design was described as 'quasi-experimental, single-group study'
Authors stated 'causality cannot be established'
Sleep outcomes were assessed only via self-report questionnaires; no objective sleep measurements were used
Authors recommend future randomized controlled trials and objective sleep measurements to validate findings
Participants were adults aged 18 to 75 years in the United States who self-reported poor sleep
What This Means
This research suggests that an AI-powered chatbot designed to improve sleep may be a feasible, easy-to-use, and acceptable tool for people who struggle with poor sleep. In the study, 107 American adults with self-reported poor sleep were asked to text with a virtual sleep therapist powered by a large language model for two weeks. About half of those who registered finished the full program, and those who did reported high satisfaction and rated the chatbot as effective. Participants spent an average of about an hour total chatting with the bot across the two weeks, in sessions averaging around nine minutes each.
Among the 42 participants included in the final analysis, sleep improved substantially across nearly all measures after using the chatbot. On average, people slept 1.4 more hours per night, fell asleep about 31 minutes faster, and had better sleep efficiency. They also reported less severe insomnia, less daytime sleepiness, better perceived sleep quality, and improved sleep hygiene habits. The only measure that did not show a statistically significant change was the sleep environment score.
This research suggests that AI chatbots built on large language models could be a low-cost, accessible way to deliver personalized sleep support at scale. However, because the study had no control group and relied entirely on self-reported sleep data, it is not possible to say the chatbot definitively caused these improvements. The authors call for future randomized controlled trials and studies using objective sleep tracking devices to confirm these promising early findings.
Liu X, Liu J. (2026). Assessing the Feasibility, Usability, Acceptability, and Efficacy of an AI Chatbot for Sleep Promotion: Quasi-Experimental Study.. JMIR formative research. https://doi.org/10.2196/84023