Write & Improve at SDU University — A Kazakhstani Classroom Trial of AI vs Teacher Writing Feedback

SDU University, a private university in Almaty, Kazakhstan, ran a quasi-experimental study comparing AI-generated writing feedback against traditional teacher feedback for first-year students in its two foreign languages (TFL) teacher-training programme. The AI tool used was Write & Improve, a free platform built by Cambridge University Press & Assessment on a 30-million-word error-annotated learner corpus, which scores writing against the CEFR and gives feedback on content, coherence, lexical resource and grammar/accuracy. Four instructors each taught one control class using traditional teacher feedback and one experimental class using Write & Improve feedback only, covering eight groups and 134 students (117 female, 17 male).

The study set out to test whether AI-generated feedback from Write & Improve could match teacher feedback in supporting trainee English teachers' writing development on IELTS-style tasks. It aimed to establish, through a controlled classroom comparison, whether the AI tool could serve as a viable, scalable supplement to human feedback in large or under-resourced classrooms.

Over five weeks from late October to late November 2024, the 134 students completed weekly IELTS-style writing tasks with pre- and post-tests, receiving either AI feedback from Write & Improve or feedback from their instructor. Researchers used Mann-Whitney U tests to confirm group equivalence at baseline, then one-way MANCOVA to compare post-test scores on task achievement, coherence and cohesion, lexical resource, and grammar and accuracy across the four teachers' paired classes.

Across all four teachers' classes and the combined sample of 134 students, the MANCOVA found no statistically significant overall difference between AI feedback and teacher feedback on any of the four writing dimensions assessed. One teacher's class showed a significant multivariate effect apparently driven by task achievement, but follow-up tests found no dimension-specific difference. Longitudinal tracking across three additional assignments showed fluctuating, inconsistent score trends in both the AI-feedback and teacher-feedback arms.

The authors concluded that Write & Improve performed as well as, but not better than, trained teacher feedback, supporting its use as a scalable supplement for large or under-resourced classrooms rather than a proven upgrade over human instruction. They noted the small, single-institution sample as a limitation, that some teachers found the tool insufficiently customisable and were reluctant to rely on it, and that the study flagged an unaddressed risk of students' lack of attention to ethical standards when using AI writing tools.