Generative Artificial Intelligence Chatbots for Motivational Interviewing: A Scoping Review From System Design to Intervention Outcomes
Runze Hu, Jingqi Kong, Yang Yang, Yihang Yang, Jingyao Liu, Haizhou Tang, Shanghang Zhang, Zheng Liu
cs.CL
Quick verdict
Scoping review of 48 GenAI-MI chatbot studies: MI-consistent output, favorable user perceptions, but behavior-change evidence thin; a credible evidence map.
Summary
This PRISMA-ScR scoping review synthesizes evidence on generative AI chatbots that deliver motivational interviewing (MI), searching nine medical and computer-science databases from January 2015 to June 2026. From 723 records, 47 reports comprising 48 studies were included and mapped across five domains: system design, safety measures, MI quality, user perceptions, and intervention outcomes. Most systems were text-based (39/48), disembodied (42/48), and English-language (40/48); 23/48 incorporated dynamic adaptation. Safety reporting was uneven: privacy and data protection (20/48) and safety-oriented content generation (17/48) predominated over automated (7/48) or human (6/48) risk monitoring. Observer-based (24 studies) and client-based (9 studies) assessments generally indicated MI-consistent interactions, whereas natural language generation metrics (5 studies) proved uninformative about MI fidelity. User perceptions, mapped onto 13 constructs across 27 studies, were favorable, especially empathy, usability, helpfulness, and intention to use, despite heterogeneous measures. Eighteen studies reported intervention outcomes, but 15 involved only a single session; range-normalized motivation changes (+2.9% to +16.0%) coexisted with sparse and mixed behavioral evidence. Key limitations: no formal risk-of-bias appraisal, English-only inclusion, and heterogeneity that precluded effect-size pooling; the review is an evidence map, not an effectiveness estimate, and is transparently reported and honestly bounded.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 2
● low 3
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- GenAI-MI chatbots can produce MI-consistent interactions, as indicated by observer-based (24 studies) and client-based (9 studies) evaluations, while NLG metrics (5 studies) fail to capture MI fidelity.
- User perceptions are generally favorable — especially empathy, usability, helpfulness, and intention to use — across 13 mapped constructs in 27 studies, though measurement is heterogeneous.
- Safety practice skews preventive: privacy/data protection (20/48) and safety-oriented content generation (17/48) far exceed automated (7/48) or human (6/48) runtime risk monitoring.
- Intervention evidence is dominated by single-session studies (15/18) showing short-term motivation gains (range-normalized +2.9% to +16.0%), with limited and mixed behavioral/functional outcomes.
- Future work needs runtime safety monitoring, standardized MI quality assessment, and longer-term comparative designs with behavioral outcomes.
Strengths
First dedicated evidence map for GenAI-MI chatbots
Extends prior AI-MI reviews (ref 11, mostly rule-based systems) by covering the design-only stage; spans the full translation continuum: design 40/48, feasibility 27/48, evaluation 18/48 (Table 2), with non-exclusive typing made explicit.
Cross-disciplinary, registered, PRISMA-ScR-compliant search
Nine databases spanning medicine (PubMed, PsycINFO, Scopus, WoS) and CS venues (ACL Anthology, arXiv, DBLP, IEEE, ACM) capture a literature that medical-only searches would miss; OSF protocol (xzw4g), dual-reviewer screening/extraction with third-reviewer arbitration.
Construct mapping makes heterogeneous perception data legible
Honest inferential discipline
Concrete field diagnosis
Weaknesses
No critical appraisal of included studies
Figure 2 invites the misreading it disclaims
Possible circularity in automated MI quality evaluation
Search boundary and registration timing
Verification burden pushed to appendices
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This PRISMA-ScR scoping review (OSF protocol xzw4g) maps 48 studies from 47 reports on generative-AI chatbots delivering motivational interviewing, retrieved from nine databases spanning medicine and computer science, January 2015 through June 2026. All included work dates from 2023 onward (3 studies in 2023, 9 in 2024, 22 in 2025, 14 in 2026). Across five research questions the paper claims: systems remain text-based (39/48), disembodied (42/48) and mostly English (40/48), with dynamic adaptation in 23/48; safety reporting privileges preventive measures — privacy/data protection (20/48), safety-oriented content generation (17/48) — over runtime safeguards (automated 7/48, human 6/48); observer-based (24 studies) and client-based (9) evaluations generally indicate MI-consistent interactions while NLG metrics (5 studies) fail to capture MI fidelity; user perceptions across 27 studies and 13 constructs are favorable; and intervention outcomes (18 studies) show short-term motivation gains (+2.9% to +16.0% range-normalized) with sparse behavioral evidence. The abstract faithfully reflects the body, and the claims are appropriately hedged throughout.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
Can AI chatbots actually deliver motivational interviewing?
According to this scoping review of 48 studies, chatbots built on generative AI produce conversations that observers and clients generally judge consistent with motivational interviewing principles, and users rate them favorably on empathy, usability, helpfulness and intention to use. However, nearly all intervention evidence comes from single sessions and short-term motivation measures; there is not yet reliable evidence these chatbots produce sustained behavior change.
How many studies were in this review and where were they found?
The authors searched nine databases across medicine and computer science (PubMed, Web of Science, Scopus, PsycINFO, DBLP, IEEE Xplore, ACM Digital Library, arXiv, ACL Anthology) and included 47 reports covering 48 studies, all published 2023 or later. Most systems were text-based, English-language, and disembodied; 23 of 48 incorporated dynamic adaptation of dialogue strategy.
What safety gaps did the review identify for GenAI mental-health chatbots?
Preventive safeguards dominate the literature: privacy/data protection (20/48 studies) and safety-oriented content generation such as restricted prompts (17/48) are far more commonly reported than runtime safeguards — automated risk monitoring in only 7/48 studies and human oversight with escalation in 6/48. Among studies with real users, 75% (21/28) reported informed consent or user education.
Do motivational interviewing chatbots change real health behavior?
Evidence is weak so far. Of 18 studies reporting intervention outcomes, 15 involved a single chatbot session and most measured proximal motivation (readiness, confidence, intention), with range-normalized gains of +2.9% to +16.0%. Behavioral outcomes were sparse and mixed: e.g., one 4-week physical activity trial raised the share meeting 150 min/week (31% to 74%) but found no significant step-count difference versus a comparator.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1 and 2 (PRISMA flow diagram; motivation-change dot plot) are described in text but not analyzable as images; individual per-study points in Figure 2 could not be verified.
- Multimedia Appendices 1–4 (PRISMA-ScR checklist, full search strategies, extraction form, extracted dataset) are referenced but not included in the provided text, so search-string quality and extraction completeness are unverifiable.
- The range-normalized change formula and some table rows show extraction artifacts; all cited numbers were cross-checked against the prose where possible.
- Several included items are 2026 preprints or in-press articles; their peer-review status could not be determined from the provided text.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.