Beyond Novelty: Longitudinal Outcomes of AI-based AWCF in Japanese EFL

by Douglas Emmett

What happens when AI is not used as a shortcut, but as part of a structured revision cycle inside the classroom? This session shares findings from a nine-month, longitudinal, quasi-experimental study of Arizona AI, a custom research platform designed to support English paragraph writing in a one-device-per-student secondary school environment. Arizona AI was built to address a familiar problem: students often do not receive timely enough feedback to revise frequently and meaningfully, while teachers do not have unlimited time for line-by-line correction. Using a digital-analog hybrid approach, students draft, receive immediate feedback, revise, and produce a final version as part of an embedded classroom routine.

The study combines scores from two blinded external raters with longitudinal surveys and learner comments. Results showed statistically significant gains for the experimental group over the comparison group, and student feedback suggests that the value of AI shifted over time from sentence-level accuracy toward stronger cohesion, organization, and idea development. Just as importantly, the project points toward a more realistic vision of classroom AI: not magic, not replacement, but a scalable way to increase meaningful revision and make better use of teacher time. The session will also introduce the latest version of the Arizona AI research platform, including customizable writing feedback modules (Peccary Paragraph) and Socratic chatbots (Desert Sage), and give participants a hands-on look at where the project is headed next.