{"id":"W4417274150","doi":"10.2196/79838","title":"Evaluating Generative AI Psychotherapy Chatbots Used by Youth: Cross-Sectional Study","year":2025,"lang":"en","type":"article","venue":"JMIR Mental Health","topic":"Digital Mental Health Interventions","field":"Psychology","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Transparency (behavior); Chatbot; Audit; Mental health; Generative grammar","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01294636,0.0003393209,0.0006299473,0.001735,0.00115699,0.001586741,0.0005785561,0.0007721176,0.00130592],"category_scores_gemma":[0.02381905,0.0006588343,0.0008527588,0.001041001,0.0008414015,0.001414229,0.001382897,0.001230751,0.0004135898],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001107499,"about_ca_system_score_gemma":0.001056215,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003661774,"about_ca_topic_score_gemma":0.006977811,"domain_scores_codex":[0.9939896,0.002749515,0.0007386832,0.0006402857,0.001396861,0.0004849834],"domain_scores_gemma":[0.9799226,0.006172695,0.006847142,0.001044383,0.00446649,0.001546646],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001655578,0.0008602614,0.9637243,0.000241211,0.0001179674,0.0001449285,0.02290726,0.00005702845,0.0003988271,0.00008978417,0.0004253456,0.01086759],"study_design_scores_gemma":[0.0000216174,0.001543827,0.9699814,0.0002024373,0.0001017743,0.0003969985,0.02512463,0.0004419317,0.000453131,0.00005747628,0.001649591,0.00002514336],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.998711,0.0001726019,0.0002555271,0.00003648701,0.000007097555,0.0002486764,0.0001588319,0.000005227772,0.0004045089],"genre_scores_gemma":[0.9975194,0.0002978545,0.0009477838,0.0001091604,0.00001163823,0.0005190535,0.0002889607,0.00000813472,0.0002979769],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01294636,"threshold_uncertainty_score":0.06846774,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1311945745376717,"score_gpt":0.5640660058849891,"score_spread":0.4328714313473174,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}