Do Students Verify What AI Tells Them? A Survey of AI Use and Self-Assessed AI Competence in High School Physics

Abstract: Generative artificial intelligence (AI) tools are increasingly used by secondary school students, yet in physics their fluent answers may contain conceptual or procedural errors. This study surveyed 480 upper-secondary students (Grades 10 - 12) in Lao Cai province, Vietnam, to describe how they use AI tools in physics learning, how they self-assess their competence in working with these tools, and what conditions and preferences they report regarding AI-integrated instruction. An online questionnaire of 18 five-point Likert items organised in three groups was administered; internal consistency was acceptable (α = .849–.914; ω = .851–.915). Students reported using AI tools occasionally (M = 3.13), mainly to preview lesson content and to obtain solutions to exercises, and less often for experimental design or data processing. Self-assessed competence was at the agreement level (M = 3.64), with 27.7% to 29.8% of students neutral on each item. At the item level, the distributions of self-assessment and reported practice did not correspond: 64.0% of students agreed that they distinguished their own work from AI-generated content, whereas 51.7% reported submitting unedited AI output at least sometimes. Most students expressed readiness to prepare before class (71.0%) and wished to be taught how to check AI answers (69.4%), while 11.5% reported insufficient devices or connectivity at home. The findings suggest that physics instruction should include explicit verification tasks and disclosure requirements, and that students’ AI competence should be assessed with evidence beyond self-report.