1 / 2026-08-15 10:25:54

Challenging AI Feedback: Questioning Patterns and Adoption Decisions in AI-Assisted IELTS Writing

AI-assisted writing feedback; challenging AI; questioning typology; selective feedback adoption; critical AI literacy

As generative AI increasingly mediates L2 writing feedback, IELTS learners rely on AI chatbots to revise their essays; yet how learners critically engage with AI feedback, which is not always accurate, remains underexplored. This exploratory study examined how learners challenge AI feedback—how often, in what forms, and how questioning shapes subsequent acceptance or rejection of AI suggestions.



Twenty-five undergraduates in an advanced IELTS course within a College English program at a Chinese university were asked to conduct at least five rounds of interaction with an AI chatbot (Doubao or DeepSeek) to revise IELTS Task 2 essays. Of the 25 students, 21 submitted assignments and 17 provided complete logs of their AI interaction. Questioning moves were iteratively coded into a six-type typology (verification, refutation, metacognitive monitoring, critique, quality audit, and assertion), and challenge-response-adoption sequences were traced.



Three findings emerged. First, challenging AI was rare: most learners passively accepted AI feedback, and none identified the technical errors we found in AI responses (e.g., a misjudged modal verb and a mislabeled clause type). Second, when challenges did occur, they took varied forms and led to negotiated outcomes of adoption, rejection, or impasse; in one striking case, a student rebutted the AI's grammar explanation yet insisted on her own view. Third, rejection reasons were typically non-technical, ranging from stylistic preference and task interpretation to, in one case, internalized instructional norms. Challenging appeared across proficiency levels, including band-5.0 writers.



Questioning frequency and type are thus measurable indicators of critical AI literacy. Scaffolded interventions—challenge scaffolds, feedback-verification training, and role-reversal review—are needed to cultivate learners' capacity to question AI critically.