New Doc on the Block: Scoping Review of AI Systems Delivering Motivational Interviewing for Health Behavior Change

Area of Science:

• Digital Health
• Artificial Intelligence
• Behavioral Science

Background:

• Artificial intelligence (AI), particularly large language models (LLMs), is increasingly integrated into digital health solutions.
• AI-driven systems are being developed to deliver motivational interviewing (MI) at scale for patient engagement and behavior change.
• Key questions persist regarding the fidelity and effectiveness of AI in replicating MI principles and achieving significant behavioral impact.

Purpose of the Study:

• To conduct a scoping review of empirical studies evaluating AI-driven systems that utilize MI techniques for health behavior change.
• To examine the feasibility, fidelity to MI principles, and reported outcomes (behavioral, psychological, engagement) of these AI systems.

Main Methods:

• Systematic search of major databases (PubMed, Embase, Scopus, Web of Science, Cochrane Library) for studies published between January 2018 and February 2025.
• Inclusion criteria focused on AI systems using natural language processing or computational logic to deliver MI for specific health behaviors.
• Data extraction by three independent reviewers on study design, AI modality, MI components, health focus, fidelity assessment, and outcomes.

Main Results:

• Fifteen studies met inclusion criteria, with most being exploratory (40%) or pilot studies (20%); only 20% were randomized controlled trials.
• AI modalities included rule-based chatbots (60%), LLM-based systems (27%), and virtual agents (13%), targeting behaviors like smoking cessation and substance use.
• High feasibility (87%) and user acceptability were reported, with moderate to high MI fidelity in 40% of studies. Users found AI non-judgmental, but limitations in empathy and safety were noted. Significant behavioral changes were reported in only 20% of studies.

Conclusions:

• AI-delivered MI holds potential for enhancing patient engagement and scaling behavior change interventions, showing early promise in usability and partial MI fidelity.
• Most current AI systems are in early development stages and lack rigorous evaluation, necessitating further research.
• Future research should prioritize randomized controlled trials, standardized MI fidelity measures, and robust safeguards for LLM safety, empathy, and accuracy in health contexts.