{"id":"W4406956279","doi":"10.2196/preprints.71434","title":"Evaluating the Usability, Technical Performance, and Accuracy of Artificial Intelligence (AI) Scribes for Primary Care: A Competitive Analysis (Preprint)","year":2025,"lang":"en","type":"preprint","venue":"","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Usability; Computer science; Data science; Artificial intelligence; World Wide Web; Human–computer interaction","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.1532872,0.0006544742,0.001187859,0.007503888,0.003015441,0.005146567,0.00139944,0.0009428693,0.001167213],"category_scores_gemma":[0.3171058,0.0003747363,0.002245634,0.006833259,0.002839419,0.00246188,0.003169758,0.0009882258,0.0002726268],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008936984,"about_ca_system_score_gemma":0.01257973,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01248274,"about_ca_topic_score_gemma":0.02050208,"domain_scores_codex":[0.8298298,0.1069614,0.02108026,0.003533381,0.03690547,0.001689729],"domain_scores_gemma":[0.424575,0.4216461,0.03784172,0.00853899,0.104966,0.00243224],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002451331,0.002626026,0.3386186,0.01514804,0.001624119,0.0003778473,0.1049662,0.002231468,0.00357224,0.003600274,0.00931214,0.5154717],"study_design_scores_gemma":[0.0007560935,0.02445481,0.7747066,0.01202419,0.002592791,0.0006021573,0.134404,0.01029924,0.01227449,0.003471664,0.02393824,0.0004757602],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9622853,0.00423396,0.01604641,0.001109485,0.0001677364,0.00881238,0.0009241043,0.0000810289,0.006339627],"genre_scores_gemma":[0.9247644,0.002160866,0.06230224,0.0005501892,0.00009224274,0.008395826,0.0007890757,0.00005239203,0.0008927485],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1532872,"threshold_uncertainty_score":0.8106697,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2253286741302461,"score_gpt":0.4998815748510081,"score_spread":0.274552900720762,"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."}}