Generative AI mediates how contested political events become publicly knowable, including when access to documentation is restricted. We conceptualise these politically structured evidentiary environments as information voids and introduce synthetic gap-filling to describe how Generative AI turns incomplete records into complete visual representations. Specifically, we examine how ChatGPT and Gemini visualised the 2014 Crimean voting procedure presented by Russia as a 'referendum', compared with the 2014 Scottish and 2017 Catalan referendums. Using repeated ambiguous prompts and manual visual content analysis, we analysed 270 images generated across private interfaces and APIs. A minimal change from 'referendum' to 'results' activated different visual schemas: voting scenes versus charts. Scotland was mostly represented as a competitive political choice (37/39 Scotland 'result' images), while Catalonia was frequently associated with police intervention and confrontation (22/45 Catalonia 'referendum' images). Crimea was predominantly represented through Russian territorial symbols (63/90 Crimea images), orderly civilian participation (40/45 Crimea 'referendum' images), and overwhelming numerical closure (all 45 Crimea 'results' images). The findings show how GenAI places politically and legally different events within a shared procedural and outcome-oriented visual grammar, and demonstrate how inequalities in the production and archiving of political knowledge can persist.
- From Information Voids to Synthetic Gap-Filling: How Generative AI Visualises Contested Political Events
- Serge Poliakoff - Stockholm School of Economics, Center for Statecraft and Strategic CommunicationJulia Kling - Swedish Defence University
- Social Science Research Network (SSRN)
- 22
- Center for Statecraft and Strategic Communication
- English
- Working paper