Interrogating ChatGPT-generated language advice: critical AI literacy as situated evaluative practice in EFL education

Generative artificial intelligence (GenAI) increasingly provides language learners with linguistic advice, yet critical AI literacy requires more than recognising that such advice may be unreliable. This study examined how critical AI literacy was enacted as EFL learners evaluated ChatGPT-generated language advice across repeated encounters. Twenty-two B1 learners in an English preparatory programme participated in a five-week qualitative repeated-task study using AI Challenge Tasks, which presented contestable advice for learners to accept, modify, reject, or investigate further. Data from task recordings, decision logs, and stimulated-recall interviews were analysed through episode-based qualitative and longitudinal comparison. Findings showed that learners differed in what they treated as open to judgement and in the grounds they established for deciding whether AI advice should influence their language choices. Acceptance, rejection, questioning, and verification were not themselves reliable indicators of critical engagement; their significance depended on the linguistic claim, available evaluative resources, and grounds supporting the decision. Across repeated encounters, practices showed change, stability, and inconsistency rather than a common developmental trajectory. The study foregrounds evaluative responsibility as an analytical lens for understanding the situated enactment of critical AI literacy and positions contestable AI use as a pedagogical approach for making ChatGPT-generated linguistic advice an object of evaluative inquiry.