The rise of AI-driven chatbots such as ChatGPT, Gemini, Le Chat, and Copilot has marked a significant shift in how people seek information online. Instead of relying solely on traditional search engines, many users are turning to these large language models (LLMs) for their ability to summarize content and provide direct links to sources, offering a seemingly faster and more efficient way to access information. However, this convenience comes with its own set of challenges and potential pitfalls.

One of the most well-known issues with AI chatbots is their propensity for making mistakes, a phenomenon known as hallucination. In other instances, they may combine information in a way that leads to incorrect conclusions, a process experts refer to as confabulation. Furthermore, these chatbots are susceptible to being influenced by false news, particularly concerning current events. This vulnerability has been highlighted by NewsGuard, a US-based company dedicated to assessing the credibility and transparency of online sources.

NewsGuard's investigations have uncovered instances where AI chatbots propagate disinformation, particularly from sources linked to Russian propaganda efforts. One notable example is the English version of the Pravda news website, which has been found to contain numerous false claims. The spread of disinformation through AI chatbots raises concerns about the potential for these platforms to be exploited for malicious purposes.

The organization Portal Kombat, identified as a key player in online disinformation, has been scrutinized for its support of the Russian army's activities in Ukraine through a network of internet portals spreading pro-Russian propaganda. VIGINUM, a French government investigation unit focused on countering foreign manipulation of information, has linked at least 193 websites to Portal Kombat between September and December 2023. The online news portal Pravda, available in multiple languages, appears to be a central component of this network.

VIGINUM's report indicates that Portal Kombat's websites primarily reproduce content from three sources: social media accounts of Russian or pro-Russian actors, Russian news agencies, and official websites of local institutions. These articles often exhibit awkward wording, transcription errors from Cyrillic to the Latin alphabet, and biased commentary. The American Sunlight Project (ASP) has identified over 180 suspected Portal Kombat sites, estimating that approximately 97 domains and subdomains within the Pravda network publish more than 10,000 articles daily. These articles cover a wide range of topics, including NATO, allied countries, and regions where Russia seeks geostrategic influence, such as the Middle East, Asia, and Africa.

NewsGuard's research suggests that the sheer volume of articles published by networks like Pravda is designed to influence the outputs of large language models (LLMs). The intention is that these models will perceive and cite these publications as legitimate sources, thereby spreading disinformation. NewsGuard's analysis of chatbot responses to false claims originating from the Pravda network revealed that in approximately one-third of cases, the chatbots confirmed pro-Russian disinformation. In less than half of the cases, they accurately presented the facts, while in the remaining instances, they declined to answer.

Another NewsGuard study, conducted in January 2025 across seven languages, found that chatbots still confirmed fake news in over a quarter of their 2,100 responses, even when not restricted to pro-Russian content. These findings underscore the ongoing challenge of preventing AI chatbots from disseminating false information.

For users of LLMs, vigilance is paramount. Given the prevalence of misinformation online, it is crucial to verify information from multiple sources. According to McKenzie Sadeghi of NewsGuard, simply rephrasing prompts is often insufficient to avoid false information. Instead, he advises users to cross-reference chatbot responses with reliable sources. However, caution is necessary, as fake news portals may mimic the layouts of established media outlets to appear credible.

Timeliness also plays a role in the accuracy of chatbot responses. Chatbots are more likely to err when dealing with new false reports that are rapidly spreading. However, over time, they are more likely to encounter fact-checks that debunk these claims, improving their accuracy.

In conclusion, while AI chatbots offer a convenient way to access and summarize information, users must remain vigilant and critically evaluate the information they receive. By checking sources, comparing responses from different chatbots, and staying informed about potential sources of disinformation, users can mitigate the risks associated with relying on AI-generated content.