How AI chatbots and AI overviews parse foreign propaganda : NPR

Key Points:

• NPR and NewsGuard conducted an experiment testing popular AI chatbots and search engine AI summaries against false narratives spread by China, Iran, and Russia, finding that AI chatbots debunked falsehoods about 75% of the time, outperforming traditional search engine results.
• AI summaries at the top of search engine results performed worse than chatbots, with significant variation between products: Google's AI Overview mostly debunked false narratives, Bing's summaries often failed, and DuckDuckGo's results were intermediate.
• Both AI chatbots and search engines cited state-aligned media sources at similar rates, but chatbots more consistently challenged false premises, while search engines sometimes only presented false information on their first page.
• Experts emphasize the importance of verifying primary sources regardless of whether research begins with AI tools or traditional search engines, noting that AI responses sometimes lack full source support or include nuanced caveats about information reliability.
• The study highlights that while AI tools are not perfect, they offer a promising starting point for investigating disinformation, especially when users critically assess sources and leverage follow-up queries to improve response accuracy.