Abstract Generative artificial intelligence (GenAI) is reshaping academic writing in higher education faster than institutions can develop evidence-informed guidance, leaving practice ahead of proof. To clarify what is happening and where benefits and risks cluster, the researchers conducted a scoping review structured by a Population–Concept–Context (PCC) frame and aligned with PRISMA-ScR procedures. Peer-reviewed, English-language empirical studies published from 2024 through Q2 2025 in higher-education settings were included, and findings were synthesized via convergent integration that juxtaposed quantitative distributions with qualitative themes. A total of 25 studies met criteria. Across populations and contexts, GenAI was most often positioned as assistive scaffolding across the planning-to-revision span of writing; reported benefits concentrated on organization, fluency, efficiency, and language support (notably for multilingual writers). Recurrent risks included hallucinations and unreliable or fabricated citations, inconsistent disclosure or attribution, and overreliance when use was unscaffolded; the limited reliability of AI-detection tools complicated integrity judgments. Context shaped practice: clearer policies and better access supported more constructive use, while the evidence base skews toward English-medium, well-resourced institutions and relies heavily on short-term or proxy outcomes. By integrating counts and themes within a PCC frame, this review offers an up-to-date evidence map that distinguishes where benefits reliably cluster (process-level supports) and where risks persist (source work and attribution), while surfacing salient gaps (faculty/postgraduate cohorts and Global South contexts). Overall, the pattern supports an assistive, not substitutive stance in which GenAI complements—rather than replaces—human judgment in argument construction, source interrogation, and synthesis.
Gabay et al. (Tue,) studied this question.