Abstract:

We conceptualise thesis writing as a sociomaterial practice in which students, supervisors, texts, material resources, and institutional norms converge in a hybrid and fluid event (Fenwick & Landri, 2012; Guerrettaz et al., 2021). Generative AI is increasingly woven into writing practices, supporting idea generation and revision while raising concerns about authorship, integrity, and supervision. However, research has rarely examined institutionally configured GPTs that operate as local, knowledge-representative co-supervisors.

This exploratory case study investigates “Dr. Artie”, a teacher-configured ChatGPT instance designed to function as an artificial co-supervisor for L1 teacher students and undergraduate thesis writers (n 15). Instruction-tuned with course guides, supervision materials, and methodological texts, the configured GPT provides formative feedback on language, structure, analytical depth, and argumentative cohesion. It offers only sentence-level examples, explicitly refusing to generate paste-ready passages, and consistently affirms student authorship and supervisor primacy. Drawing on sociomaterial and actor–network-informed perspectives (cf. Barad, 2007), we conceptualise students, supervisors, the GPT instance, and institutional policies as an assemblage of distributed agencies, and examine how this assemblage mediates writing practices and reconfigures supervision and responsibility.

The data consist of fieldnotes, prompt logs from the configuration of Dr. Artie, and open-ended survey responses in which students describe their uses of, and orientations to, the co-supervisor. Inspired by materials-in-use approaches (Guerrettaz et al., 2021), the analysis traces chains of intra-action through which prompts, interface features, course documents, and human actors become aligned in producing particular textual moves and evaluative framings. In the paper session, we present preliminary findings on how agency is distributed, assembled, and potentially destabilised, and how it shapes the trajectories of student–GPT–supervisor–policy entanglements. The study contributes to writing research by exploring and theorising a configured GPT as a sociomaterial, agentic co-supervisor rather than merely a tool in academic text production.

Keywords:

Sociomateriality, thesis writing, generative AI, supervision.