AI use, perceived writing needs, and emerging changes in student academic writing in ESP higher education

Abstract

Generative AI (genAI) is increasingly becoming part of how students approach academic writing, yet its impact does not unfold evenly across all aspects of the writing process. Much of the existing research highlights improvements in accuracy and organisation, but the question of how writing changes at the level of discourse remains less explored. This study takes that perspective by combining questionnaire data (n = 354) with an analysis of bachelor thesis introductions (n = 60) produced in an ESP context. The analysis draws on Ken Hyland’s (2010) metadiscourse framework and compares two sets of texts written in 2020–2021 and 2024–2025, reflecting different writing conditions in relation to genAI availability. What emerges is not a uniform improvement, but a more uneven pattern. Students tend to perceive their writing as clearer and easier to produce, and this is visible in the texts through less explicit structural signalling and more frequent clarification and referencing practices. At the same time, features connected to stance and epistemic positioning show little variation. Even where students recognise possible risks such as overreliance or reduced independent thinking, this awareness does not consistently shape how arguments are constructed. This tension is captured through the notion of asymmetrical development, understood as uneven progress across different layers of writing. By placing student perceptions alongside discourse-level evidence, the study offers a more differentiated account of genAI in academic writing and points to the need for more explicit attention to stance, argumentation, and critical engagement in ESP instruction.