{"id":"W4413101237","doi":"10.1177/12034754251364852","title":"Assessing the Accuracy, Quality, and Inclusivity of AI Chatbots for Patient Education on Psoriasis: A Year-Long Comparative Study","year":2025,"lang":"en","type":"letter","venue":"Journal of Cutaneous Medicine and Surgery","topic":"Mobile Health and mHealth Applications","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University; McGill University Health Centre; Université de Montréal","funders":"","keywords":"Medicine; Psoriasis; Quality (philosophy); Quality of life (healthcare); Dermatology; Medical physics; Medical education; Nursing","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02575586,0.0003084627,0.0007371486,0.002008389,0.001233207,0.002008315,0.001248831,0.002030713,0.002121046],"category_scores_gemma":[0.1446318,0.0005006985,0.001102522,0.001451882,0.001205281,0.003630366,0.001393108,0.001475046,0.001099993],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001750432,"about_ca_system_score_gemma":0.001084654,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008348647,"about_ca_topic_score_gemma":0.008850278,"domain_scores_codex":[0.9819478,0.0080207,0.002768886,0.001178943,0.005489654,0.0005940721],"domain_scores_gemma":[0.7649627,0.1542579,0.03670442,0.01025494,0.02967331,0.004146578],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006387799,0.003620317,0.9120163,0.0004556794,0.0005960856,0.0002620053,0.01790929,0.0002554107,0.0009188571,0.0001196729,0.0009865003,0.0564722],"study_design_scores_gemma":[0.0003371905,0.01335301,0.9693325,0.0002586743,0.0006271908,0.0005567474,0.009594432,0.001679749,0.001215877,0.0001465695,0.002789018,0.0001091202],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9970789,0.0003317541,0.0002698709,0.0002632101,0.00004639084,0.000122336,0.0001932364,0.00001457683,0.001679838],"genre_scores_gemma":[0.9971404,0.0002959222,0.0006183413,0.0003063174,0.00006904348,0.0001323088,0.0002233327,0.00001670143,0.00119756],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9742441,"threshold_uncertainty_score":0.1362116,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1572203867822606,"score_gpt":0.5291544534937564,"score_spread":0.3719340667114959,"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."}}