MétaCan
Menu
Back to cohort
Record W4417274150 · doi:10.2196/79838

Evaluating Generative AI Psychotherapy Chatbots Used by Youth: Cross-Sectional Study

2025· article· en· W4417274150 on OpenAlexvenueno aff
Kunmi Sobowale, D. Humphrey, Sophia Yingruo Zhao

Bibliographic record

VenueJMIR Mental Health · 2025
Typearticle
Languageen
FieldPsychology
TopicDigital Mental Health Interventions
Canadian institutionsnot available
Fundersnot available
KeywordsTransparency (behavior)ChatbotAuditMental healthGenerative grammar

Abstract

fetched live from OpenAlex

Background: Many youth rely on direct-to-consumer generative artificial intelligence (GenAI) chatbots for mental health support, yet the quality of the psychotherapeutic capabilities of these chatbots is understudied. Objective: This study aimed to comprehensively evaluate and compare the quality of widely used GenAI chatbots with psychotherapeutic capabilities using the Conversational Agent for Psychotherapy Evaluation II (CAPE-II) framework. Methods: In this cross-sectional study, trained raters used the CAPE-II framework to rate the quality of 5 chatbots from GenAI platforms widely used by youth. Trained raters role-played as youth using personas of youth with mental health challenges to prompt chatbots, facilitating conversations. Chatbot responses were generated from August to October 2024. The primary outcomes were rated scores in 9 sections. The proportion of high-quality ratings (binary rating of 1) across each section was compared between chatbots using Bonferroni-corrected chi-square tests. Results: While GenAI chatbots were found to be accessible (104/120 high-quality ratings, 86.7%) and avoid harmful statements and misinformation (71/80, 89%), they performed poorly in their therapeutic approach (14/45, 31%) and their ability to monitor and assess risk (31/80, 39%). Privacy policies were difficult to understand, and information on chatbot model training and knowledge was unavailable, resulting in low scores. Bonferroni-corrected chi-square tests showed statistically significant differences in chatbot quality in the background, therapeutic approach, and monitoring and risk evaluation sections. Qualitatively, raters perceived most chatbots as having strong conversational abilities but found them plagued by various issues, including fabricated content and poor handling of crisis situations. Conclusions: Direct-to-consumer GenAI chatbots are unsafe for the millions of youth who use them. While they demonstrate strengths in accessibility and conversational capabilities, they pose unacceptable risks through improper crisis handling and a lack of transparency regarding privacy and model training. Immediate reforms, including the use of standardized audits of quality, such as the CAPE-II framework, are needed. These findings provide actionable targets for platforms, regulators, and policymakers to protect youth seeking mental health support.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.013
metaresearch head score (Gemma)0.024
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Observational · Consensus signal: Observational
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.013
Threshold uncertainty score0.068

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0130.024
Meta-epidemiology (narrow)0.0000.001
Meta-epidemiology (broad)0.0010.001
Bibliometrics0.0020.001
Science and technology studies0.0010.001
Scholarly communication0.0020.001
Open science0.0010.001
Research integrity0.0010.001
Insufficient payload (model declined to judge)0.0010.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.131
GPT teacher head0.564
Teacher spread0.433 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designObservational
Domainnot available
GenreEmpirical

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations12
Published2025
Admission routes1
Has abstractyes

Explore more

Same venueJMIR Mental HealthSame topicDigital Mental Health InterventionsFrench-language works237,207