Designing Reasoning-Rich Tasks for ChatGPT-Assisted Academic Writing: A Qualitative Classroom-Based Study
DOI:
https://doi.org/10.36312/jllet.v1i2.74Keywords:
ChatGPT, Academic writing, Task-based language teaching, Critical reasoning, Generative artificial intelligenceAbstract
The increasing use of ChatGPT in academic writing raises an important pedagogical question concerning the conditions under which generative artificial intelligence promotes reasoning rather than encourages cognitive outsourcing. This qualitative classroom-based study investigated how Task-Based Language Teaching (TBLT) task characteristics shaped learners' reasoning and interaction with ChatGPT and identified task conditions and forms of teacher mediation that promoted critical engagement. Participants were 21 EFL learners from private educational institutions in Bangladesh who had experience with online learning, academic writing, and ChatGPT. Data were collected through semi-structured interviews and analyzed using thematic analysis combining deductive and inductive coding. The findings revealed a continuum between reasoning-poor and reasoning-rich task conditions. Descriptive, summarizing, grammar-focused, and reproduction-oriented tasks generally encouraged transactional interaction, including direct requests for drafts, paraphrases, and corrections. Conversely, argumentative, comparative, problem-based, source-based, and revision-oriented tasks promoted dialogic interaction characterized by follow-up questioning, counterargument generation, evidence verification, comparison, and explicit justification. Critical engagement was further strengthened by teacher-mediated verification requirements, justification of AI-related decisions, modeling of critical AI evaluation, dialogic questioning, structured comparison, peer discussion, and guided autonomy. The findings indicate that ChatGPT does not inherently strengthen or weaken reasoning; its cognitive value depends substantially on task architecture and instructional mediation. The study concludes that reasoning-rich TBLT should preserve learners' epistemic responsibility by requiring them to interrogate, verify, compare, revise, and justify AI-generated content. These findings provide practical implications for designing AI-assisted academic writing activities that promote critical reasoning rather than passive dependence.






