The Institute for Strategic Dialogue (ISD) has researched the most popular chatbots for Russian propaganda about the war in Ukraine — ChatGPT, Gemini, Grok, and the Chinese DeepSeek. The research was conducted in 5 languages: English, Spanish, French, German, and Italian.
Chatbots cited links to Russian state sources in almost 25% of cases.
The Institute for Strategic Dialogue (ISD) has researched the most popular chatbots for Russian propaganda about the war in Ukraine — ChatGPT, Gemini, Grok, and the Chinese DeepSeek. The research was conducted in 5 languages: English, Spanish, French, German, and Italian.
Chatbots cited links to Russian state sources in almost 25% of cases.
ISD tested 300 queries and received content linked to Russian state sources in varying amounts, depending on the neutrality of the query itself, The Register writes .
When researchers entered neutral queries, Russian state content appeared about 11% of the time; for biased queries, the figure was 18%; and for malicious (manipulative) queries, the figure was 24%.
According to the study, nearly 25% of malicious queries aimed at returning pro-Russian information contained sources attributed to the Kremlin, compared to just 10% for neutral queries. The researchers suggest that this indicates the possibility of manipulating LLMs to bias them towards positions supported by Russian state media.
ChatGPT provided Russian sources almost three times more often for malicious queries compared to neutral ones.
Grok cited roughly the same number of Russian sources for each query category, indicating that wording matters less for this model.
DeepSeek produced 13 links to state media, with biased queries yielding more pro-Kremlin content than manipulative ones.
Gemini , which included the fewest government sources, showed only two links to neutral queries and three to malicious ones.
The ISD study also found that the language used for the requests did not have a significant impact on the likelihood that LLMs would express pro-Russian views.
