Conspiracy and Populist Bias in AI: A Comparative Analysis of Chatbots

Populism and conspiracy theories are mutually reinforcing phenomena, as both construct narratives that oppose "the people" to allegedly corrupt elites, thereby fostering institutional distrust. Artificial intelligence (AI) may intensify this dynamic by disseminating polarizing content, reinforcing echo chambers, and amplifying conspiracy-driven populist rhetoric through generative tools. AI chatbots, with their human-like conversational style, can make the information they present appear more credible, potentially reinforcing users' existing beliefs or even changing their views. The rapid proliferation of AI chatbots, coupled with a lack of transparency and accountability in their underlying algorithms, may contribute to the spread of anti-pluralist ideas and distorted narratives (Bontridder & Poullet, 2021). At the same time, AI systems also hold potential as instruments to counteract radical ideologies and conspiracy theories (Montoro-Montarroso et al., 2023; Costello et al., 2024). This paper proposes a new multidimensional methodology to assess how AI chatbots respond to conspiratorial narratives and evaluate populist leaders. The approach combines AI-on-AI analyses such as ideological scaling, sentiment analysis, and topic modelling. Results from a comparative analysis of ChatGPT, Grok, Copilot, DeepSeek, Gemini, and Perplexity reveal that, while chatbots largely reject conspiracy theories (e.g., the Great Replacement, Deep State, COVID-19 denial), they exhibit notable biases and discrepancies in their evaluations of populist leaders’ discourses and policies, especially when dealing with some countries that are not very salient in Western media.