Russian Propaganda Found in One-Fifth of AI Chatbot Responses on Ukraine, Study Reveals

Russian propaganda appears in approximately one in five artificial intelligence chatbot answers about the Ukraine conflict, according to a comprehensive new study by the British think tank Institute of Strategic Dialogue (ISD).

Researchers posed more than 300 questions to leading AI platforms—OpenAI’s ChatGPT, Google’s Gemini, xAI’s Grok, and Deepseek’s V3.2—in five different languages, using neutral, biased, and deliberately malicious phrasing to analyze response patterns.

The findings reveal a troubling tendency for AI systems to exhibit “confirmation bias,” with chatbots mirroring users’ language choices and subsequently drawing from questionable sources. Russian propaganda was particularly prevalent when questions contained loaded language or misleading premises.

“When users frame questions in certain ways, these AI systems appear primed to respond in kind, often by pulling from problematic information sources,” said one of the researchers involved in the study, whose name wasn’t specified in the report.

The investigation found that OpenAI’s ChatGPT provided three times more Russian sources when responding to biased or malicious prompts compared to neutral questions about the conflict. Meanwhile, Grok delivered the highest number of Russian-backed sources even when faced with neutrally worded queries.

Each platform demonstrated distinct vulnerabilities. In two specific instances, Deepseek provided four links to Russian-backed sources in a single response—the highest concentration observed in the study. These sources included VT Foreign Policy, which the report identifies as a distributor of content from known Russian propaganda operations such as Storm-1516 and the Foundation to Battle Injustice, alongside state media outlets Sputnik and Russia Today.

Grok showed a concerning tendency to quote Russia Today journalists directly, often linking to their social media posts as authoritative sources. The researchers noted this approach “blurs the lines between overt propaganda and personal opinion” and raises serious questions about AI systems’ ability to identify sanctioned state media content when it appears via third-party channels.

Google’s Gemini demonstrated the strongest safeguards, refusing to engage with some maliciously framed prompts and instead delivering warnings about “inappropriate or unsafe” content. However, researchers criticized Gemini for often failing to disclose its information sources, making verification challenging.

The study uncovered specific topic areas where Russian sources appeared most frequently. Questions about Ukraine’s military recruitment practices were particularly problematic, with 40 percent of Grok’s responses and over 28 percent of ChatGPT’s answers citing at least one Russian source. Both platforms provided Kremlin sources in 28.5 percent of their responses to various Ukraine-related queries.

Conversely, questions about war crimes and Ukrainian refugees triggered fewer Russian-backed sources across all four platforms, suggesting more robust information ecosystems around these topics.

The researchers theorize that AI systems gravitate toward Russian sources when confronting what experts call “data voids”—search terms lacking high-quality information from mainstream sources. As the American think tank Data and Society has documented, these voids typically emerge around obscure topics or breaking news situations where credible journalism hasn’t yet established a footprint.

The findings come amid growing concern about AI systems’ vulnerability to manipulation and the spread of disinformation, particularly regarding geopolitical conflicts. Technology companies face mounting pressure to implement more rigorous source verification and build stronger safeguards against the amplification of state propaganda.

The study underscores the complex challenges facing AI developers as they attempt to create systems that provide accurate information while remaining responsive to user queries—a balance that appears increasingly difficult to maintain in contentious geopolitical contexts.