Exploring human mobility discourses through LLMs

This work was presented at the conference Digital Humans Today: Critical Inquiry with and about the Digital (King’s College London), in the session "Platform Urbanism and Mobility". We investigate how ChatGPT and Gemini infer user context from IP geolocation. Using identical prompts submitted from the origin and destination countries of an established migration corridor — in this case, the Brazil–Portugal corridor — the study examines how these models silently adapt their responses according to the user’s inferred location.

By holding prompts constant and varying only location, the research isolates geolocation as a variable and surfaces the often-invisible ways these systems construct the user. The work contributes to a critical inquiry into the digital by treating LLMs not as neutral tools but as actors that participate in the production of mobility discourses, with implications for how migrant populations are represented and addressed.