The Bias in the Machine: How Language and Location Shape Chatbot Accounts of the Israel–Gaza Conflict

Chatbots are increasingly used as everyday sources of information. However, their outputs vary depending on language, geography, and opaque platform policies, making them prone to various biases. This, combined with users’ tendency to over-trust these systems, poses significant risks. These concerns are particularly acute in conflict environments, where disinformation is widespread, and narratives are highly contested. Yet, despite their growing influence, there is limited empirical evidence on how chatbots operate in such high-stakes contexts. This study addresses this gap through an exploratory algorithmic audit examining chatbot responses to politically sensitive prompts related to the Israel–Gaza conflict. Using a structured, prompt-based comparison across three major chatbot platforms and three language-region pairings (Hebrew–Israel, Arabic–Lebanon and Israel, English–the UK), the analysis reveals notable differences in how chatbots present facts, attribute responsibility, and select information sources. The findings also reveal systematic patterns in how chatbots prioritize information, such as a tendency to rely heavily on official state institutions, even when those sources circulate contested or misleading claims. These discrepancies illustrate how AI systems can generate and reinforce divergent informational realities within conflict settings, potentially entrenching competing narratives, deepening polarization, and undermining prospects for shared understanding. The study underscores the need to critically examine chatbots as emerging actors in international politics, shaping how knowledge, credibility, and responsibility are constructed during armed conflict.