ChatGPT’s interpretation of contested memoryscapes favours the voice of current governments, capitalism and the far-right

Generative AI, such as ChatGPT-4o, exhibits a notable bias toward dominant political and economic narratives when interpreting contested memoryscapes linked to dictatorial regimes. A study analyzing responses about three Spanish-speaking memorial sites—Valley of the Fallen (Spain), ESMA Museum (Argentina), and Monument to Salvador Allende (Chile)—revealed a preference for perspectives aligned with current governments, capitalism, and far-right rhetoric, particularly in Spanish-language outputs. Using least-to-most prompting and Discourse-Historical Approach (DHA), researchers demonstrated how AI can inadvertently reinforce power structures, posing challenges for unbiased heritage narratives. This finding underscores the need to critically assess AI's role in shaping memory politics and democratic discourse.

González-Cantera, T., & Bonacchi, C. (2025). ChatGPT’s interpretation of contested memoryscapes favours the voice of current governments, capitalism and the far-right. International Journal of Heritage Studies, 31(11), 1508–1530. https://doi.org/10.1080/13527258.2025.2543736