Structured AI workflows in Saudi EMI research writing

This mixed-methods case study examines a staged integration of generative AI (GenAI) into a core English course, Research Writing Techniques at a private English-medium university in Saudi Arabia (English-Medium Instruction; EMI) that enrolls predominantly first-year Arabic L1 students. Across three consecutive semesters, the intervention progressed from AI-enhanced instructor materials (Semester 241), to optional student-facing support via a course-specific CustomGPT tutor and short recap videos (Semester 242), to full Week-1 integration of a mandatory homework pipeline in which students consulted the course Guide, practiced with the CustomGPT tutor, reviewed via Quizlet, and completed proctored quizzes through Examplify, alongside a curricular shift in Assignment 1 from an Annotated Bibliography to a Literature Review (Semester 251). Data sources included ExamSoft midterm and final exam scores, quiz gradebooks, aggregate CustomGPT usage metrics, YouTube analytics, student surveys, and brief email interviews. Compared to baseline, Semester 251 students scored higher on homework—even under stricter, proctored conditions (76.00% vs. 94.61%; Hedges’ g = 1.20) and moderate gains on comprehensive final exams (61.36% vs. 70.46%; g = 0.52). A large midterm effect (64.29% vs. 85.01%; g = 1.19) and reduced score variability indicate that the structured workflow promoted deep, more equitable learning in APA citation and research methods.