Tracing Artificial Intelligence in EFL Student Writing: A Linguistic Approach to Identifying Machine-Assisted Text in Undergraduate Writing

Abstract

As AI-assisted writing tools become more present in university classrooms, the nature of students’ written work in EFL contexts is quietly shifting. These developments raise new questions about who is responsible for the final text, how students build their writing abilities, and how they develop a sense of voice in their academic work. Although interest in the ethical and pedagogical implications of AI is increasing, we still know little about how and where AI-shaped textual patterns actually appear in student writing. This study investigates the linguistic and stylistic traces of AI assistance in undergraduate academic work and highlights the need for fair, transparent assessment practices and informed pedagogical support in AI-integrated learning environments. Using a qualitative corpus of 557 undergraduate texts, the study applied a three-cycle coding process supported by descriptive linguistic measures. This approach offers a detailed view of textual patterns by focusing on the features found in students’ actual writing, moving beyond the self-report surveys and tool-detection methods commonly used in earlier research. The findings reveal recurring tendencies, including reduced sentence-length variation, over-explicit cohesion, generic macrostructural framing, and a polished yet impersonal tone. These tendencies point to a mixed effect: AI tools seem to improve cohesion and linguistic accuracy, yet they can also blur students’ individual voice and weaken the texture of their arguments. The study offers evidence that can help educators and institutions make sense of AI’s growing role in student writing and develop assessment and teaching practices that fit their pedagogical and institutional realities.