Large language models are increasingly used for learning and decision-making, yet little empirical work has examined how they respond to geographically ambiguous prompts that do not specify a national context. This study examines whether AI-generated responses default to United States–specific institutional frameworks and how such defaults operate for users in Canada. Using a prompt-based experimental design, 72 geographically ambiguous prompts spanning ten structural domains and one values-based domain were administered five times each using the free version of ChatGPT(ChatGPT 5.2- 5 Mini) under typical access conditions, including observed transitions between available free-tier models, yielding 360 responses.Analysis shows a clear divergence between institutional framing and value expression. While responses to values-based prompts consistently emphasized communitarian and public-oriented principles familiar to Canadian users, structurally framed prompts frequently relied on U.S.-specific laws, agencies, and procedures presented without qualification. This combination appears to increase user trust while simultaneously introducing foreign institutional assumptions, making national defaulting more difficult to notice in practice.By documenting this pattern, the study contributes to research on digital colonization by identifying institutional defaulting as a distinct form of epistemic influence. The findings have implications for education, digital sovereignty, and AI governance, particularly for students and other users who rely on free-tier AI systems for everyday informational and institutional guidance.
Karen Elizabeth Hamilton (Fri,) studied this question.