Trabajo aceptado para presentar en junio 2026 de modo presencial. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . l. . . . . . . . The transition to AI-assisted writing in postgraduate classrooms highlights the risk of thesis writers delegating knowledge construction to AI (Kosmyna et al., 2025), undermining authorial agency and writing's epistemic potential. This paper identifies pedagogical conditions for a critical integration of Gemini at the end of an 18-month Thesis Writing Seminar within a Master’s program for educators. The assignment aimed to improve the literature review section in advanced drafts of thesis proposals (Carlino et al, 2026). We examine Gemini interaction records, resulting proposal drafts, and the accompanying forum. Despite frustrations with Gemini’s source retrieval, the process fostered critical AI literacy, enabling students to refine their drafts through strategic collaboration. On the one hand, this process encouraged a critical stance toward machine-generated outputs, highlighting the imperative for conceptual and terminological precision when guiding search queries. On the other hand, two pedagogical conditions were essential for students to effectively leverage AI interactions. First, the Gemini task was assigned at the end of the seminar, after students had developed knowledge and experience to monitor GAI interactions. Second, the iterative writing of the proposal drafts throughout the seminar fostered a sense of intellectual ownership over their work. This made students less likely to relinquish their authorial voice. These conditions empowered students to prompt the AI and discern novel intertextual connections between the literature review and with their specific research objectives. To prevent students from delegating their thinking or uncritically accepting GAI-generated outputs, certain pedagogical conditions must be established prior to GAI integration. Specifically, students must first engage in iterative drafting, making deliberate compositional decisions while internalizing textualization criteria and rhetorical purposes. This didactic framework empowers students to engage in epistemic interactions with AI, enabling them to discern where GenAI can provide meaningful contributions (Martínez, 2025; Tate, 2025). References Carlino, P., Calderón, L., Roni, C., & Rosli, N. (2026). Taller de proyecto de tesis: enseñar escritura académica desde el enfoque de producción auténtica. Praxis Educativa, 30 (1), 1-20. https://doi.org/10.19137/praxiseducativa-2026-300105 Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X. H., Beresnitzky, A. V., Braunstein, I., y Maes, P. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. MIT Media Lab. Martínez, N. V. (2025). Escritura, escuela, IA y enseñanza: Revoluciones, expectativas y desengaños. Textos. Didáctica de la Lengua y de la Literatura, (108), 31-36. Tate, T. P., Harnick-Shapiro, B., Ritchie, D. R., Tseng, W., Dennin, M., y Warschauer, M. (2025). Incorporating generative AI into a writing-intensive undergraduate course without off-loading learning. Discover Computing, 28(72). https://doi.org/10.1007/s10791-025-09563-9.