This study explores students’ perceptions of using DeepL Translate in translating Arabic assignments. The increasing adoption of AI-based translation tools among students offers convenience and efficiency, yet raises issues of accuracy, contextual understanding, and academic dependency. Using a qualitative phenomenological approach, five active students of the Arabic Language Education Study Program at IAIN Kendari were purposively selected as participants. Data were collected through in-depth semi-structured interviews and analyzed thematically. The findings reveal three major perception patterns: (1) critical use, where two students utilized DeepL as an initial aid while revising outputs manually; (2) pragmatic dependence, where two others relied entirely on the tool due to limited mastery of nahwu and sharaf; and (3) reflective evaluation, where one participant verified translations through peer discussion and comparison with other sources. Although all participants acknowledged DeepL’s speed and ease of use, they also noted frequent inaccuracies in translating Islamic and scientific texts. The study concludes that DeepL can enhance comprehension and efficiency when used reflectively, but pedagogical guidance is essential to strengthen linguistic awareness, translation competence, and ethical responsibility in AI-assisted translation.