Artificial Intelligence–Based Chatbots for Promoting Health Behavioral Changes: Systematic Review
Journal of Medical Internet Research · 5 authors, 4 centres
AI SUMMARY
FIDELITY 92%
POPULATIONAdults and adolescents from diverse clinical and general populations (e.g., office workers, smokers, patients with obesity, breast cancer, substance misuse, Medicare recipients); sample sizes ranged from 20 to 99,217.
INTERVENTIONAI-based chatbots delivering behavior change interventions (e.g., goal setting, monitoring, real-time feedback, on-demand support) using techniques such as NLP, ML, and hybrid algorithms.
COMPARISONVaried: usual care, pharmaceutical treatment alone, app without chatbot, or no control group (most studies were pre-post designs).
This summary was generated by AI from a single paper. It has not been reviewed by a clinician and is not clinical advice. Verify against the source before acting on it.
This systematic review of 15 studies found that AI-based chatbots are efficacious in promoting healthy lifestyles (40% of studies), smoking cessation (27%), medication adherence (13%), and reducing substance misuse (7%). While chatbots offered personalized, nonjudgmental, and scalable interventions, results for feasibility, acceptability, and usability were mixed, and most studies had moderate-to-high risk of bias. The findings support the potential of AI chatbots for health behavior change but highlight the need for more rigorous randomized controlled trials.
Full summary
4,035 CHARS
**Background:** Artificial intelligence (AI)–based chatbots offer personalized, engaging, and on-demand health promotion interventions. Prior reviews focused on mental health or limited behaviors and did not distinguish AI-driven from rule-based chatbots. This review aimed to evaluate feasibility, efficacy, and intervention characteristics of AI chatbots for a wide range of health behavior changes.
**Methods:** A comprehensive search was conducted in June 2022 across 7 databases (PubMed, IEEE Xplore, ACM Digital Library, PsycINFO, Web of Science, Embase, JMIR publications) for empirical articles published from 1980 to 2022. Inclusion criteria: intervention research on health behaviors; empirical studies using chatbots; chatbots built on AI platforms or algorithms (ML, deep learning, NLP, natural language understanding); studies reporting qualitative or quantitative results; English articles. Exclusion criteria: non-full-text articles; rule-based chatbots; unclear AI algorithms; studies focused only on mental health. Of 1,961 initially retrieved articles, 15 met inclusion criteria. Data extraction and quality assessment were performed independently by two authors using the NIH quality assessment tool for controlled intervention studies and the CONSORT-AI extension.
**Key Results:** The 15 included studies were published between 2011 and 2021 (40% from 2019). Sample sizes ranged from 20 to 99,217 (median 116). Most studies were conducted in high-income countries (31% US, 15% Australia). Target behaviors: healthy lifestyle (physical activity/diet; 33%), smoking cessation (27%), treatment/medication adherence (20%), reducing substance misuse (7%). Only 27% (4/15) were RCTs; 60% used pre-post designs.
EFFICACY
For healthy lifestyles, 4/5 studies reported increased physical activity (e.g., +109.8 minutes, P=.005; 3.58 times higher total physical activity, P<.001). Weight loss averaged 2.38% in one study. Diet improved (31% increase in healthy meals; Mediterranean diet scores +5.7 points, P<.001). For smoking cessation, the intervention group had 2.44 times greater odds of abstinence at 1 month (P<.001); one study reported 28.9% completed cessation goals (10% higher than without chatbot). For substance misuse, significant increases in confidence to resist urges (+16.9, P<.001) and decreases in substance use occasions (−9.3, P<.001). For medication adherence, 17.4% of refill reminders resulted in actual refill requests; adherence improved >20% in 4 weeks.
FEASIBILITY
73% of studies reported feasibility metrics. Engagement decreased over time (retention from 72% at month 2 to 31% at month 8). Acceptability: satisfaction was <50% in most studies, though 93.95% satisfaction was reported in one study. Chatbots provided a nonjudgmental safe space for sensitive disclosures. Usability: mixed—content was generally reliable and understandable, but technical issues (82.3% reported problems in one study) and impractical recommendations were noted.
QUALITY ASSESSMENT
Risk of bias was moderate to high. Only 33% of studies had low dropout rates; 33% reported power calculations; 40% used intent-to-treat analysis. No studies reported concealment of assigned intervention from evaluators. AI quality assessment: rationale for AI was specified in all studies, but only 7% described input data characteristics and handling of unavailable data.
**Clinical Implications:** AI chatbots demonstrate potential for scalable, personalized, and nonjudgmental health behavior interventions. They can be integrated into existing platforms (smartphones, Facebook Messenger) and offer 24/7 support. However, the evidence base is limited by methodological weaknesses, lack of standardized outcome measures, insufficient description of AI techniques, and limited generalizability to low-income countries and younger populations. Future research should prioritize robust RCTs, standardized reporting of AI algorithms, assessment of safety and ethics, and studies in diverse geographic and demographic settings.
PICO
PPOPULATION
Adults and adolescents from diverse clinical and general populations (e.g., office workers, smokers, patients with obesity, breast cancer, substance misuse, Medicare recipients); sample sizes ranged from 20 to 99,217.
IINTERVENTION
AI-based chatbots delivering behavior change interventions (e.g., goal setting, monitoring, real-time feedback, on-demand support) using techniques such as NLP, ML, and hybrid algorithms.
OOUTCOME
Primary: health behavior change (physical activity, diet, smoking cessation, medication adherence, substance use reduction). Secondary: feasibility, acceptability, usability, engagement.