Generative AI Use in Professional Graduate Thesis Writing: Adoption, Perceived Outcomes, and the Role of a Research-Specialized Agent
Survey of 83 MBA students reveals near-universal AI adoption with 77% heavy use and 6.27/7 perceived quality improvement.
A new study by researchers Kenji Saito, Rei Tajika, and Satoru Shibuya reveals near-universal adoption of generative AI among graduate students writing professional theses. Surveying 83 MBA students in Japan (36.1% response rate), the research found that 95.2% reported at least some AI use, with 77.1% classified as heavy users. Students integrated AI across the entire research-writing workflow—from literature review and drafting to consultation when stuck—reporting significant benefits including clearer argument structure (82.3%), better revision quality (73.4%), and faster writing (70.9%). The mean perceived quality improvement was 6.27 out of 7.
Despite these gains, concerns persisted about output accuracy (75.9%) and proper citation handling. The study also tested GAMER PAT, a research-specialized AI agent, against general AI tools. Preferences significantly favored GAMER PAT for inquiry deepening and structural organization (both p < 0.05). Follow-up interviews revealed students employ active "epistemic vigilance" strategies and differentiate tool use across thesis phases. The central implication is that the educational challenge has shifted from adoption to verification, source governance, and specialized AI tool design, with GAMER PAT demonstrating that research-focused scaffolding provides measurable advantages.
- 95.2% of surveyed MBA students used AI for thesis writing, with 77.1% as heavy users reporting 6.27/7 quality improvement
- Students preferred research-specialized agent GAMER PAT over general AI for inquiry deepening and structural organization (p < 0.05)
- Despite benefits, 75.9% expressed concerns about output accuracy, highlighting need for verification and specialized tool design
Why It Matters
Shows AI is already deeply embedded in academic work, forcing education to focus on verification and specialized tool design over basic adoption.