Teachers' Barriers in Utilising Artificial Intelligence in Learning: A Phenomenological Study at an Islamic Primary School

Abstract

Artificial intelligence (AI) is increasingly entering Indonesian teaching practice, yet recent national evidence suggests that teachers' technical knowledge and classroom integration remain uneven despite generally positive attitudes. This study examined barriers experienced by teachers in utilising AI in learning at SD IT Darul Fikri, North Bengkulu, and explored how those barriers were experienced and interpreted. A qualitative study with a descriptive phenomenological orientation was conducted. Sixteen classroom and subject teachers served as the primary participants, while an ICT technician, an administrative staff member, the principal, and two foundation managers provided contextual information for triangulation. Data were collected through lesson observation, semi-structured interviews, and documentation. Interview data were organised into significant statements, meaning units, sub-themes, and a synthesis of essential meanings; contextual data were used to corroborate conditions surrounding teachers' experiences. The analysis identified three interrelated barrier categories: competence barriers, infrastructure barriers, and technical-institutional support barriers. Teachers described difficulty formulating precise prompts, basic feature mastery, the need to verify AI outputs, uneven readiness, unstable network coverage, insufficient or unevenly distributed devices, incidental ICT assistance, uncertainty in selecting suitable tools, and a gap between institutional policy and routine classroom use. Across the teacher accounts, four essential meanings were identified: barriers were treated as learning challenges rather than reasons to stop; AI was understood as a support tool rather than a substitute for pedagogical judgement; self-directed learning developed in response to limited mentoring; and verification of AI output was understood as professional responsibility. The findings suggest that where basic AI-supporting infrastructure has been introduced, practical integration may still be constrained by capacity, distribution, competence, and institutionalised support.