While chatbots frequently identified faulty premises in misleading questions, the analysis revealed that AI summaries—generated answers appearing at the top of search engine results—performed less reliably, often failing to challenge disinformation at a higher rate than traditional search links.
These findings matter because they highlight a shift in how AI infrastructure handles the "poisoning" of information by foreign influence campaigns.
While researchers feared that generative AI would uncritically repeat state-sponsored falsehoods, the chatbots often outperformed traditional web searches by analyzing the credibility of sources and synthesizing debunks across multiple languages.
However, the inconsistent performance of search-based AI summaries suggests a risk for users who rely on the automated snippets at the top of Google or Bing results, where the mechanism for grounding answers in search data can sometimes lead to the amplification of unverified claims.
Moving forward, digital literacy experts suggest that a diversified search strategy is essential as AI models continue to integrate into daily internet use.
To improve accuracy, users can ask chatbots to "take a second look" at evidence or prompt them to evaluate the specific credibility of cited sources.
While companies like Google and Microsoft continue to refine these tools, the study warns that the quality of AI-generated answers remains tied to the availability of factual reporting; when reliable sources are scarce, the systems are more likely to surface inaccurate information.