Multilingual students in English medium instruction contexts develop academic English under conditions that shape learning and identity. This study examines how such students experience AI-mediated interaction as a site for multilingual meaning making, what affective responses accompany these practices, and how students negotiate academic identity and legitimacy, with a focus on how this emerging technology functions as an innovation in language learning. Drawing on interpretative phenomenological analysis, the study explores the lived experiences of 15 multilingual students at an English-medium university in Northern Cyprus using ChatGPT and Gemini. Data were generated through AI diaries and semi-structured interviews, capturing participants' practices as they occurred in context. Findings suggest that students use GenAI as a translanguaging space that supports understanding and rehearsal of academic English while maintaining expectations of English-only performance. Building on participants' experiences within this EMI context, the study proposes the notion of a conditional AI-mediated translanguaging space. Participants report affective ambivalence, as confidence coexists with guilt and uncertainty, reflecting the moral dimensions of engaging with innovation under institutional constraint. The study highlights the need for instructional approaches that acknowledge multilingual processes and provide clearer guidance on AI use, and argues that innovation in language learning is context-dependent and mediated through existing pedagogical and institutional frameworks rather than inherently transformative.
