An experiment conducted by NPR shows that AI-based chatbots, in identifying and rejecting misleading narratives published by entities linked to China, Iran and Russia, overall performed better than traditional search engines. However, these tools still face challenges such as source quality and accuracy of information.
In this experiment, researchers designed 30 questions based on false narratives that were published between December 2025 and July 2026. These questions were given to chatbots including ChatGPT and Gemini as well as to several major search engines.
Chatbots were more successful at debunking false narratives
According to the results of this review, AI chatbots were able to identify and reject misleading narratives in about three-quarters of cases. In many instances these tools first recognized the incorrect assumption underlying the question and then provided corrected information.
For example, one of the questions related to a Russian narrative about the bombing of a historic monastery in Kyiv. The chatbots corrected the claim and explained that the site was targeted during a Russian airstrike and the narrative blaming Ukraine was part of a misinformation campaign.
Search engines performed differently
By contrast, AI-generated summaries in the top results of Google, Bing and DuckDuckGo performed worse than the chatbots.
Google’s “AI Overview” feature was able to reject more false narratives, but Bing’s AI summaries failed in many cases. DuckDuckGo’s performance fell between those two search engines.
The experiment’s results also showed that chatbots and search engines cite links to media outlets controlled by or aligned with governments at roughly similar rates; a point that again highlights the importance of source quality.
Question wording and language can change the AI response
Another important finding of the review shows that how a question is phrased and the language used can affect the responses produced by AI tools.
Google, criticizing the study’s methodology, said the questions used in the experiment do not necessarily represent how users typically use a search engine. Microsoft, meanwhile, said its services rely on search results and users should review and verify the sources provided.
AI is not yet a replacement for independent verification
Despite the chatbots’ better performance in this experiment, researchers and experts stress that AI tools still cannot replace independent verification and fact-checking.
In this context, a study from the University of Washington showed that about one-ninth of the real claims made in Google’s AI-based summaries are not fully supported by the sources cited for them.
Overall, NPR concluded that AI tools can, in some circumstances, be a more effective tool for countering foreign propaganda and misleading narratives than traditional search methods; however, the credibility of responses still depends on the quality of information available on the internet and users’ ability to check and verify primary sources.