1 Introduction

This section introduces the importance of writing feedback in second language teaching and its drain on teacher resources, noting that ChatGPT, as a large language model, demonstrates potential in providing deep-level writing feedback, capable of transforming traditional feedback methods and enhancing efficiency. It further proposes that this study aims to evaluate ChatGPT's ability to correct errors and provide feedback in Chinese second language writing, offering a reference for teachers considering the adoption of this tool.

2 Literature Review

This section reviews relevant research foundations across three dimensions. Regarding factors influencing Chinese second language writing quality, accuracy, diversity, and complexity have been verified as core linguistic elements, among which lexical complexity, correctness, and grammatical accuracy demonstrate the strongest predictive power for writing quality; meanwhile, content quality and discourse features are particularly critical for advanced-level learners, yet in teaching practice, teacher feedback often emphasizes linguistic form while neglecting the content dimension. Concerning feedback effectiveness, the superiority of direct versus indirect correction remains inconclusive, with feedback effects depending on learners' subjective perceptions and the quality of teacher-student interaction; suggestive comments enjoy higher acceptance due to their operability. At the ChatGPT application level, existing research indicates that it can provide balanced feedback across language, content, and structure, with advantages including identifying strengths, offering improvement plans, and enhancing learning motivation; however, limitations include inaccurate feedback, information overload, unclear categorization, lack of specificity, and unstable consistency. Therefore, researchers generally agree that it can only serve as a teacher assistance tool, requiring integration with professional teacher judgment to correct deficiencies.

3 Research Questions

Based on existing research gaps, three core research questions are proposed: comparing the characteristic differences between Chinese teachers' and ChatGPT's writing feedback, evaluating the accuracy and effectiveness of ChatGPT's feedback across content, linguistic form, and structure, and investigating Chinese teachers' attitudes toward the application of ChatGPT-assisted feedback.

4 Research Design

This section introduces a mixed-methods approach to compare the similarities and differences between teacher and ChatGPT feedback on Chinese second language compositions. The study selected two lower-intermediate scoring advanced-level compositions from the HSK Dynamic Composition Corpus (one argumentative and one narrative), inviting five experienced Chinese teachers from different regions to provide feedback, while simultaneously using GPT-4o to generate corresponding feedback. Data collection was completed through four steps: first, designing role instructions based on scoring criteria for ChatGPT to generate feedback; second, having teachers provide feedback in their customary manner; third, conducting quantitative and qualitative comparisons across three dimensions—feedback quantity, type, and consistency—with feedback types encompassing direct feedback, informational feedback, inquiry, suggestions, and affirmation and encouragement; finally, employing a Likert scale for teachers to evaluate ChatGPT feedback's accuracy, effectiveness, and application potential, and collecting detailed opinions through interviews, with repeated verification to ensure data reliability.

5 Research Findings

This section compares the differences between ChatGPT and five Chinese teachers' feedback on two compositions, and investigates teachers' evaluations of ChatGPT feedback. In terms of feedback quantity, teachers identified significantly more linguistic errors than ChatGPT, while ChatGPT provided more feedback on content structure and overall quality than all teachers. Regarding feedback types, teachers employed various methods including direct correction, informational prompts, and clarification inquiries, whereas ChatGPT exclusively used direct modifications lacking diversity; ChatGPT's feedback structure was more systematic, covering five aspects—language, content, structure, highlights, and improvement suggestions—while teacher feedback lacked clear categorization. In consistency, approximately 79% of ChatGPT's linguistic feedback reached consensus with the majority of teachers, but consistency in content and structure was only 29%. In efficiency, ChatGPT significantly outperformed teachers' average time of 31 minutes. Teacher evaluations showed that ChatGPT received high ratings for accuracy, but showed divergence in effectiveness: acknowledging its strengths in identifying highlights and providing ideas, while criticizing its lack of specific explanations, obvious errors, absence of instructional guidance, excessive linguistic difficulty, inability to annotate, and loss of handwriting practice opportunities. Regarding ChatGPT's potential as an assistant, three teachers held positive attitudes while two were reserved, believing its feedback was difficult to use directly and might not necessarily reduce time costs.

6 Discussion

This section discusses the three research questions. First, ChatGPT and Chinese teachers differ in feedback emphasis: teachers focus on linguistic accuracy and tend toward direct correction, reflecting their primary positioning of second language writing training as linguistic training, while also exposing challenges teachers face in multi-dimensional feedback; ChatGPT provides balanced feedback across language, content, and structure dimensions, demonstrating systematicity and balance. Second, ChatGPT feedback shows relatively high accuracy, with 79% consistency with teachers in linguistic feedback and 80% consensus in content and structure, yet its instructional effectiveness is questioned: corrective feedback lacks specific explanations and guidance, while content and structure feedback is verbose and hollow with above-level terminology; the study notes that direct correction aligns with Chinese language characteristics, but ChatGPT fails to achieve the higher-order goal of guiding learners toward autonomous reflection, and problems may stem from insufficient instruction design, requiring optimization before further judgment. Third, teachers hold positive yet cautious attitudes toward introducing ChatGPT: teachers emphasizing direct correction affirm its performance, while those valuing knowledge expansion and content depth adopt cautious attitudes, with attitudinal differences closely related to their understanding of feedback purposes.

7 Conclusion

This study found that ChatGPT demonstrates advantages in Chinese second language writing feedback, including rapid text analysis, error identification, content modification, and multi-dimensional evaluation, but shows deficiencies in specific corrective information and guiding autonomous reflection. Teachers' attitudes toward it are clearly divided; ChatGPT can assist teachers in improving efficiency and expanding feedback ideas, but requires selective use based on student circumstances, and its objectivity can also reduce teacher subjective bias. Improving feedback quality requires optimizing instruction design, providing learner information, teaching objectives, and more. Study limitations include small sample size, lack of refined instruction testing, absence of domestic model comparisons, and lack of learner perspectives; future research can explore these aspects in depth.