For the final paper in this session at the 2026 International Communication Association conference in Cape Town we are going back to Mykola Makhortykh, whose focus is on the role of generative AI in representing history-related information. The vast volumes of historical information mean that AI is increasingly used to process such materials, and in recent years there has been a considerable increase in end-users engaging with AI chatbots to explore historical information.
But LLMs also generate new textual and visual content, which can make historical material more accessible but also raises questions about the fabrication of facts and information. This paper tests this in the context of how the history and memory of the Holocaust, especially with respect to Ukraine, are being represented by AI systems.
The paper conducted an AI audit with a focus on ChatGPT, Google Bard (now Gemini), and Bing Copilot, in English, Ukrainian, and Russian in September 2023, and prompted these from a range of perspectives using 333 distinct prompts. It tested whether the chat agent responds at all, whether the answer matches the prompt baseline, whether instrumentalisation of the past by Russia was mentioned, and whether false details were provided.
There was less than 50% agreement between chatbots and expert opinion for all but two of the chatbot / language combinations (for Bard in English and Ukrainian); some responses aligned more with Russian propaganda than with the accepted truth, in fact. However, Bard also stands out for providing additional false details in more than 50% of cases across all three languages; for the other chatbots that percentage is below 20%. Accuracy generally decreased for the non-English languages.
Notably, none of the chatbots mentioned the instrumentalisation of history by Kremlin propaganda. There is therefore a high risk of users being misled by these chatbots.
