AI-use orientations and self-perceived AI competence among medical students: a cross-sectional study, Brazil

Artificial intelligence (AI) is increasingly integrated into medical education, but most studies emphasize frequency of use, attitudes, or familiarity rather than the pedagogical pattern of students' interaction with AI. Evidence from Latin America, particularly northeastern Brazil, remains limited. We examined AI-use orientations and their associations with self-perceived technical-operational and critical-ethical AI competence among medical students in a predominantly Ceará-based sample. We conducted a cross-sectional study using electronic survey data collected between August and October 2025. The Artificial Intelligence Education Use instrument (AIEd-U) assessed Targeted, Supported, Enabled, and Ethical AI orientations. Participants were assigned to a predominant orientation only when one domain had a uniquely highest score; 172 of 844 respondents (20.4%) had tied maxima and were excluded, leaving 672 participants. AI competence was assessed using a study-specific nine-item questionnaire with technical-operational (Domain 1) and critical-ethical (Domain 2) dimensions. Analyses included present-sample reliability, Spearman correlations with 3,000-resample bootstrap 95% confidence intervals, and Kruskal–Wallis tests with ordinal eta-squared, Dunn-Bonferroni comparisons, and Cliff's delta. Among 672 participants, 94.5% were from Ceará. Predominant orientations were Ethical AI (44.0%), Supported AI (36.8%), Enabled AI (15.6%), and Targeted AI (3.6%). Enabled AI was most strongly correlated with Domain 1 (rho = 0.487, 95% CI 0.423–0.545), whereas Ethical AI was most strongly correlated with Domain 2 (rho = 0.437, 95% CI 0.372–0.500). Targeted AI was inversely associated with Domain 2 (rho = -0.200, 95% CI -0.270 to -0.123). Competence differed across predominant orientations for Domain 1 (H = 36.315, P < .001, ordinal eta-squared = 0.050, 95% CI 0.024–0.091) and Domain 2 (H = 26.102, P < .001, ordinal eta-squared = 0.035, 95% CI 0.015–0.063), indicating small omnibus effects. Distinct patterns of AI engagement were associated with different profiles of self-perceived AI competence. These cross-sectional findings do not establish causality, but they support further evaluation of curricula that combine practical AI use with critical appraisal, ethical awareness, and learner self-regulation.