GenAI as scholarly ally: patterns, pedagogy, and policies in graduate writing research

The integration of generative artificial intelligence into graduate-level academic writing has accelerated since 2019, yet systematic evidence regarding adoption patterns remains scattered. This study synthesises findings from 71 empirical studies identified through Scopus AI analysis, examining how Master's and PhD students employ these technologies across disciplines. Meta-analytic results indicate that students who engage in recursive interaction with AI tools achieve stronger writing outcomes (Hedges' g = 0.89, 95% CI [0.78, 1.00]) than those who use AI tools merely as editorial assistants (g = 0.41, 95% CI [0.29, 0.53]). Writing interventions that combine self-regulated learning strategies with AI support yield the largest effects (g = 0.94). Adoption rates vary by field: 78% in STEM, 71% in social sciences, and 62% in humanities, with distinct usage patterns emerging within each discipline. Analysis of institutional policies reveals that frameworks permitting structured AI use while requiring disclosure correlate with lower rates of academic misconduct (r = -0.62) compared to prohibitive approaches. These findings suggest that the effectiveness of AI in academic writing depends less on the technology itself than on the pedagogical context and strategic frameworks within which it operates. The evidence points toward a need for discipline-specific guidelines that acknowledge both the affordances and limitations of current AI capabilities.