ChatGPT use in higher education writing and its relationship with critical thinking autonomy struggle and authorship

Generative artificial intelligence (GenAI) is increasingly shaping higher education, particularly in academic writing and learning efficiency. Yet empirical evidence on its effects on students’ critical thinking, learning agency, and authorship perceptions remains limited. This study examines how ChatGPT influences undergraduate academic writing, with particular attention to critical thinking dispositions, including epistemic autonomy, reflective judgment, persistence in problem solving, and responsibility for authorship. Using a convergent mixed-methods design, the study integrated survey data from 270 students across two Indonesian universities with semi-structured interviews involving 15 purposively selected participants. Quantitative data were analyzed using descriptive statistics, reliability testing, independent-samples t-tests, and Pearson correlation analysis, while qualitative interview data were examined through thematic analysis. Quantitative findings showed generally positive perceptions of GenAI literacy and digital readiness, with students reporting particularly favorable views of efficiency, perceived writing quality, academic confidence, and learning enjoyment. Cognitive engagement was positive but less pronounced, whereas behavioral engagement, especially in collaborative activities, was less evident. Despite increased confidence, Students also expressed concerns about originality, authorship, and potential over-reliance on AI. Qualitative findings revealed three recurring mechanisms: creeping dependency, erosion of cognitive struggle, and shifting notions of authorship. These findings informed a three-part model of AI-mediated learning consisting of acceleration, attenuation, and ambivalence. Overall, GenAI enhances task efficiency while reshaping critical thinking dispositions. Limitations include the focus on two universities and a small interview sample, which may affect generalizability. Future research should examine diverse contexts and longitudinal impact of AI on higher-order learning.