English Medium Instruction (EMI) is increasingly used in higher education to promote English proficiency while delivering academic content. However, students with limited confidence in academic English often face persistent language challenges. In response, generative AI tools such as ChatGPT are being adopted to support writing and communication in EMI contexts. While these tools offer benefits, they also raise concerns about authorship, identity, and linguistic ownership.

This study begins in April 2025 in a Global Semiotics course at a Japanese university, with approximately 40 undergraduate students. The class includes both domestic (Japanese) and international students, mainly from Thailand and Malaysia, along with a few students from diverse international backgrounds, such as Vietnam, China, Korea, Europe, and the United States. Students are asked to use ChatGPT as part of a structured framework based on Bloom’s Taxonomy, which provides clear guidance on when AI may be used, while allowing flexibility in how students engage with it. The study investigates how students’ English proficiency levels relate to their approaches to AI use within this guided but flexible framework, and how these approaches reflect their perceptions of linguistic agency, which refers to their sense of control over their language use and expression.

Language difficulties in EMI are well known, but the use of AI adds a new challenge. Students may feel that the ideas or wording are not their own or become unsure about authorship. Although AI provides linguistic support, it relies on purposeful and specific input from users. Without this, it does not generate meaningful or contextually appropriate ideas. This study considers whether students recognize this and how it influences their sense of ownership over the content they produce.

A mixed-methods approach is used. Students’ English proficiency levels will be determined through placement test scores. Reflexive Thematic Analysis will be applied to reflections, written tasks, and interviews. The study is informed by principles of causal inference to explore how different proficiency levels correspond to students’ AI use strategies and the degree of authorship they experience. Preliminary findings will be available at the time of the conference.

As AI tools become more accessible in higher education, understanding their influence on learner autonomy, authorship, and identity is increasingly important. These technologies are not neutral—they shape how students think, write, and express themselves. This study offers insights for educators working in international and multilingual university classrooms, highlighting the need for teaching practices that help students engage with AI critically and ethically. It emphasizes the importance of guiding learners to reflect on authorship, take ownership of their language use, and develop the awareness needed to navigate AI-integrated academic environments. The findings also aim to inform how educators can support international students by designing AI-integrated tasks that promote critical awareness, ethical engagement, and a stronger sense of authorship in academic work.

Keywords: Generative Artificial Intelligence (AI), English Medium Instruction (EMI), Linguistic Agency, Causal Inference, Reflexive Thematic Analysis (RTA).