Published Paper Details:

This narrative review examines the intersection of large language models (LLMs) and academic writing in higher education, with a specific focus on Claude AI a model whose institutional adoption has expanded rapidly but whose representation in the peer-reviewed literature remains comparatively limited. Drawing on peer-reviewed sources published between 2020 and 2025 across databases including Scopus, Web of Science, ERIC, ScienceDirect, and Frontiers, this review synthesizes evidence on the benefits, risks, and cognitive implications of generative AI-assisted academic writing. The literature confirms that generative AI tools, led overwhelmingly by research on ChatGPT, improve measurable dimensions of student writing including cohesion, fluency, and revision behavior, while simultaneously introducing well-documented risks: over-reliance, academic integrity violations, and detector bias against non-native English writers. Evidence on cognitive and critical-thinking outcomes remains genuinely contested, with reviews reporting both enhancement and erosion of higher-order thinking depending on task design and instructional scaffolding. Against this backdrop, the review identifies a central and consequential gap: despite Claude's formal deployment across universities including Northeastern University, the London School of Economics and Political Science, and Champlain College, and despite substantial documented student and educator usage, only a small number of peer-reviewed studies examine Claude specifically within academic writing contexts. This asymmetry between institutional deployment and independent scholarly scrutiny means that current pedagogical guidance on "AI-assisted writing" rests on evidence base derived almost entirely from a different model family. The review concludes by outlining research gaps and future directions specific to Claude AI, particularly within English for Academic Purposes and English for Specific Purposes contexts, and offers implications for educators, curriculum designers, and institutional policymakers navigating this evolving and unevenly evidenced landscape.