Assessed academic writing outcomes in an AI-permitted research methodology course: evidence from EFL university students in Palestine

Abstract The rapid integration of generative artificial intelligence (AI) into academic writing has intensified debates about learning, assessment validity, and academic literacy in second and foreign language (EFL) higher education. Although an expanding literature documents students’ perceptions of AI-assisted writing, fewer studies examine assessed writing outcomes within authentic course ecologies where AI use is permitted. This study investigates assessed academic writing outcomes among 34 fourth-year EFL university students enrolled in a Research Methodology course in Palestine; 33 complete paired cases were included in the quantitative analyses. Using a longitudinal, assessment-grounded design, I compare pre-course diagnostic writing with end-of-course qualitative research papers produced under AI-permitted conditions. Writing was evaluated using analytic criteria aligned with core dimensions of academic literacy, including argument quality, vocabulary and style, grammatical accuracy, citation practices, and engagement with sources. Quantitative analyses indicated no statistically significant change in mean performance across the semester on a common percentage metric (with a slightly lower post-course mean), alongside reduced dispersion in end-of-course scores and heterogeneous individual trajectories. A rubric-informed review of the final papers nevertheless suggested comparatively stronger end-of-course patterns in argumentation, lexical control, and source engagement, alongside continued variability in grammatical accuracy and critical citation practices. Interpretively, these results are bounded by task nonequivalence (timed diagnostic versus a multi-week research paper with access to sources and permitted supports) and by the absence of process evidence on AI engagement (no usage logs or systematic self-reports), which prevents estimating the extent or type of AI assistance and rules out causal attribution. The study contributes context-sensitive evidence to debates on generative AI, academic literacy, and assessment in EFL higher education, underscoring the need for pedagogical and assessment frameworks that foreground transparent, learning-oriented engagement with AI tools.