Estonian Study: AI Feedback Helps Writing, But Overuse Risks Learning

Researchers Andres Karjus, Janika Leoste, and Tiia Õun conducted a study on repeated AI feedback in higher education, analyzing 2,988 essay-feedback-appraisal instances from 283 Estonian bachelor students over one semester. Students used self-selected AI tools with a uniform prompt to obtain feedback on reflective essays, then rated its helpfulness. The results showed that students found the AI feedback helpful and actionable more often than not, but about 10% considered it unhelpful, with dissatisfaction increasing toward the end of the semester. The study also analyzed essay reflection depth and employed a validated AI text classifier to estimate the proportion of essays likely written without significant AI aid.

The findings indicate that while generative AI provides a fast and scalable way to deliver immediate writing advice, it is not a self-contained solution for improving reflective depth. The researchers emphasize that benefits depend on whether students learn to use AI selectively and critically, without over-reliance that could harm the learning process. This study contributes descriptive classroom evidence on the integration of AI feedback, highlighting both its potential and its limitations in educational settings.