Original title: Revising with Humans and AI: Effects of Teacher-, GenAI-, and Blended-Feedback on EFL Writing Outcomes
Feifei Han, Zehua Wang · October 2026
Researchers tested three ways of getting writing feedback: from a human teacher, from generative AI, or from a mix of both. Over six weeks, 109 college students revised their essays using one of these methods. The AI helped students produce much better immediate revisions than teacher feedback. But when students had to write an entirely new essay from scratch, the teacher-guided group scored the highest. Fixing an old draft is very different from learning how to write a new one.
You are learning to cook and keep burning your sauces. If someone stands beside you and quickly adjusts the flame or adds water every time you slip up, that evening's dinner turns out delicious. But you never actually had to figure out why the pan got too hot in the first place. Here, AI feedback acted like that hands-on helper who rescues the immediate dish, while human teacher feedback taught the underlying principles needed to cook a totally new meal alone.
Students and schools are rushing to use generative AI tools to grade and fix homework drafts. This study reveals a critical trade-off: while AI helps students fix their current draft faster, learning from a human teacher makes students significantly better at writing brand new essays on their own. Knowing this helps educators avoid relying purely on automated grading.
The study was conducted over only six weeks with 109 first-year university students across three intact College English classes in China. It only evaluated writing performance, motivation, self-efficacy, and perceptions on specific narrative and revision tasks, finding no significant impact on motivation or self-efficacy.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Revising with Humans and AI: Effects of Teacher-, GenAI-, and Blended-Feedback on EFL Writing Outcomes — Feifei Han, Zehua Wang — Behavioral Sciences — doi:10.3390/bs16101837
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.