From AI Assistance to Independent Writing

A Mixed-Methods Study among Undergraduate Social Science Students

Abstract

Generative artificial intelligence (GenAI) is increasingly embedded in university students’ academic writing practices, yet it remains unclear whether improved AI-assisted texts reflect genuine development in independent writing competence. This study investigated the relationship between GenAI-assisted writing performance, independent writing competence, and students’ patterns of AI use among undergraduate social science students. An explanatory sequential mixed-methods design was employed involving 175 students from higher education institutions across Sulawesi, Indonesia, followed by qualitative interviews with 17 purposively selected participants. Quantitative data were collected through parallel AI-assisted and independent argumentative writing tasks and the GenAI Writing Strategy Questionnaire, while qualitative data were obtained through semi-structured stimulated-recall interviews. The results showed that students performed significantly better in the GenAI-assisted condition than in the independent-writing condition, revealing a substantial performance–competence gap. Critical evaluation and verification positively predicted independent writing performance and were associated with a smaller gap, whereas substitutive reliance negatively predicted independent competence and emerged as the strongest predictor of a larger gap. Qualitative findings distinguished two dominant patterns: productive augmentation, in which students retained control over argumentation and evaluation, and substitutive delegation, in which core cognitive processes were transferred to AI. The findings suggest that GenAI can improve textual quality without necessarily strengthening writer development. Therefore, academic writing pedagogy should prioritize evaluative judgment, verification, and cognitive responsibility rather than focusing solely on access to or frequency of GenAI use.