Patterns of AI-assisted academic writing of public administration students in an Indonesian university
Bevaola Kusumasari, Sajida SajidaPurpose
This article examines how undergraduate public administration students used and evaluated Artificial Intelligence (AI) tools while completing an academic writing assignment, interpreting the documented micro-level behaviors and attitudes through the Behavioral Public Administration (BPA) perspective.
Design/methodology/approach
A bounded, single-class descriptive case study was conducted with 40 undergraduate public administration students in Indonesia. Students prepared a three-page paper on the digital divide using AI-assisted tools and submitted accounts of their tools, prompts, and experiences. The analysis was based on evidence-linked qualitative coding, participant-level matrices, and descriptive counts; no control group, pre-test/post-test, or performance-effect comparison was conducted.
Findings
Based on the coding, the study has identified 65 student-tool combinations across the 40 submissions, as students often used multiple tools. ChatGPT dominated, accounting for 53.8 percent of combinations, while students also used systems and supporting digital tools for source retrieval, paraphrasing, editing, and citation work. Five distinct prompt types emerged — task-oriented, inquiry-based, sequential, context-rich, and multi-tool — revealing varied delegation strategies. Overall, 37 students expressed predominantly positive perceptions, 26 expressed negative perceptions, and 26 expressed both, often simultaneously. While speed, accessibility, and writing assistance were valued, these benefits were overshadowed by inaccurate citations, misinterpretation of prompts, dependency, plagiarism, and loss of voice, highlighting the conditional nature of students’ reliance on AI.
Originality/value
This article positions AI-assisted academic writing as a behavioral issue in public administration education, using BPA as an analytical framework rather than course content or a learning outcome․ The findings connect tool choice‚ delegation strategies‚ prompting behaviors, verification practices, trust, and ethical judgement to practical guidance on disclosure requirements, source checking, and process-based assessment․