{"id":"W4414078894","doi":"10.1016/j.cont.2025.102239","title":"315 - Reliability of AI Chatbots in Providing Urinary Tract Infection Health Information: A Comparative Study of ChatGPT, Google Gemini, and DeepSeek","year":2025,"lang":"en","type":"article","venue":"Continence","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"NOSM University","funders":"","keywords":"Reliability (semiconductor); Urinary system; Chatbot; Health care; Health services","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":[],"consensus_categories":[],"category_scores_codex":[0.07530316,0.0004664164,0.001007245,0.002095882,0.0007115239,0.002221323,0.0009911705,0.0009137475,0.001271077],"category_scores_gemma":[0.2615925,0.0007243076,0.001336316,0.001421192,0.002019134,0.002399003,0.002350571,0.001151324,0.0007496204],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001237681,"about_ca_system_score_gemma":0.001249146,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002287478,"about_ca_topic_score_gemma":0.002611604,"domain_scores_codex":[0.92516,0.04631783,0.007942487,0.005075229,0.01389188,0.001612589],"domain_scores_gemma":[0.6422718,0.2569261,0.04288957,0.01152464,0.0424743,0.003913497],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002320016,0.0008197303,0.888124,0.001492111,0.0006515117,0.0001899011,0.06410851,0.000482967,0.001315186,0.0002444698,0.0008344266,0.03941728],"study_design_scores_gemma":[0.000131934,0.003654013,0.962595,0.0006698241,0.0003572853,0.0004032368,0.02313907,0.00455064,0.001500907,0.0002825836,0.002589201,0.0001263469],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9955063,0.0003083365,0.001608207,0.00009264264,0.0000618933,0.0003305855,0.000167363,0.00004488108,0.001879629],"genre_scores_gemma":[0.9968197,0.0002135682,0.001796627,0.00008139265,0.00003783024,0.0004494618,0.0002447287,0.0000318245,0.0003249186],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07530316,"threshold_uncertainty_score":0.3982459,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09354104672687147,"score_gpt":0.4443054551261905,"score_spread":0.350764408399319,"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."}}