New study reveals why students reject ChatGPT feedback in academic writing

Generative AI tools like ChatGPT present extraordinary potential in promoting better language learning and teaching practice, such as timely feedback, ease of use, personalized support, simulations of the learning environment, and individualized recommendations on learning materials. However, a team of researchers in Hong Kong and Macao found that 45.9% of ChatGPT-generated feedback was rejected by second-language students, higher rejection rates for content-focused feedback (58.7%) than form-focused (41.3%).

To investigate why second-language learners reject feedback generated by ChatGPT. A team of researchers in Hong Kong and Macao, led by Associate Professor Wei Wei, from Macao Polytechnic University analyzed 45 Computer Science undergraduates tasked with revising argumentative essays using ChatGPT feedback. This study was published online on January 7, 2025, in the ECNU Review of Education.  Through mixed-method analysis of self-reflection and actual revision behaviors, the team identified four key factors driving rejection:

1. Mismatched expectations: Feedback often misinterpreted student intent or lacked clarity.
2. High workload: Overwhelming or vague suggestions discouraged engagement.
3. Conflicts with external references: Discrepancies between AI advice and teacher/peer feedback caused distrust.
4. Impeding conditions: Limited emotional support and non-personalized responses hindered usability.

Notably, content-focused feedback (e.g., argument structure, evidence quality) faced higher rejection due to its subjective nature, while form-focused feedback (e.g., grammar, vocabulary) was more readily accepted but still challenged by effort-related barriers. 

“Our findings reveal that students aren’t dismissing AI feedback outright - they’re struggling with its relevance and practicality,” explains Prof. Wei. “For content-related suggestions, misalignment with their goals or external standards often leads to rejection. Meanwhile, even straightforward grammar tips can feel burdensome if not contextualized.”

The study also highlights a critical paradox: while ChatGPT’s accessibility is praised, its inability to provide emotionally supportive or proficiency-tailored feedback limits its effectiveness. “Students need more than raw suggestions; they require scaffolding to interpret and apply AI-generated insights,” Prof. Wei adds.

***

Reference

Titles of original papers: Unpacking the Rejection of L2 Students Toward ChatGPT-Generated

Feedback: An Explanatory Research

Journal: ECNU Review of Education

DOI: 10.1177/20965311241305140