ChatGPT is often depicted either as an all-powerful tutor or as a threat to students' cognitive growth—both views ignoring the complex nature of learning. Under what conditions can teachers use ChatGPT critically and productively? We conducted an exploratory workshop with 147 Primary Education students over five weeks (10 hours total). Guided by revised Bloom's Taxonomy, each group completed iterative phases while interacting with ChatGPT 4 to collect information, refine texts, and design teaching scenarios. A mixed-methods design combined phase-by-phase satisfaction items with qualitative thematic analysis of metacognitive diaries (≈19,500 words) and group submissions. Qualitative evidence shows a progression from information retrieval and text refinement toward analysis and evaluation: students increasingly questioned AI outputs, triangulated sources, and justified accept/reject decisions. Quantitatively, self-reported satisfaction was high in early phases and dipped when tasks demanded higher-order reasoning, consistent with increased cognitive demands. Students learned to use ChatGPT as a collaborative tool supporting their own cognitive development rather than as an "oracle", demonstrating the importance of pedagogical scaffolding in AI-mediated learning.
Ranieri et al. (2026) studied this question.