As artificial intelligence (AI) systems become increasingly common in the creation of scientific texts, researchers have turned their attention to comparing the linguistic traits of AI-generated writing with those produced by humans. Nevertheless, some language features crucial to writing quality have received little attention. To address this gap, the present study investigates the use of metadiscourse markers in the abstract sections of research articles written by humans versus those generated by AI, while considering the influence of academic discipline. A dataset of 160 abstracts was collected: 40 from applied linguistics and 40 from mechanical engineering, drawn from reputable journals published between 2021 and 2022. For each human-written abstract in both fields, an AI-generated version was produced using a consistent prompt, resulting in 40 AI abstracts per discipline. Metadiscourse markers were identified based on Hyland’s (2005) framework, and the data were analyzed through descriptive statistics and Chi-square tests. The findings showed that AI-generated abstracts display a highly structured style with excessive use of frame markers but almost entirely lack authorial presence, evident in the near absence of self-mentions. Human authors, in contrast, achieved a refined balance of interpersonal engagement and scholarly humility through the careful use of hedges and self-mentions. Moreover, clear disciplinary differences emerged: abstracts from mechanical engineering tended to adopt a persuasive and assertive tone, using many boosters and evidentials, while those from applied linguistics followed an interpretive and organizational pattern. The study concludes that although AI can be useful for academic writing, it currently falls short of replicating the subtle interpersonal dynamics and discipline-specific rhetorical conventions essential for expert-level scholarly communication. These findings highlight that a thorough comparison of human and AI writing must simultaneously consider linguistic features, scientific genre, and disciplinary context, an approach that can inform future research in academic discourse analysis and the evaluation of AI-generated texts.
