LLMs Distort Conflict Information, Vulnerable to GEO Warfare

Artificial Intelligence (AI) answer engines, powered by Large Language Models (LLMs), are increasingly becoming a primary source of information on global conflicts for analysts, scholars, and the public. This research investigates whether these LLMs exhibit discernible patterns of error when queried about conflicts and what this implies for the global conflict information environment. The study involved asking five leading AI answer engines a battery of questions about 28 conflicts, then scoring their 5,460 responses against documented evidence. A key finding was that the thinner the retrievable record surrounding a conflict, the more prone the engines were to inventing, misattributing, or miscounting facts. This lack of robust information not only encourages hallucination but also creates a structural vulnerability to mis- and disinformation, as these "thin records" are easier to manipulate through "Generative Engine Optimization" (GEO). An analysis of over 1,000 websites used by these LLMs for conflict facts confirmed that GEO source optimization is already occurring, with state-partisan digital capture showing rapid growth. The paper discusses the implications for scholarship in an era of GEO information warfare and advocates for a renewed focus on deep local monitoring and translation-based research, which AI tools cannot replicate. It concludes by outlining future research opportunities and challenges in this rapidly evolving space.