As voters seeking information about political issues increasingly turn to LLMs to do research, will AI lead them astray with hallucinations and bias, distorting the democratic process? Maybe not, according to a study of AI use in the 2024 UK election. Lennart Luettgau, Hannah Kirk, Kobi Hackenburg, and colleagues asked UK residents, reached online in two waves of 1,711 and 1,147 participants, to research true or false claims on two of four topics: climate change, immigration, criminal justice, and COVID-19 policy, with the two un-researched topics serving as within-subject controls. Participants were randomly assigned to use GPT-4o, Claude-3.5, Mistral, or Google search. Both before and after the research phase, participants answered questions about all four topics. Research increased belief in true information and decreased belief in misinformation, no matter which method was used. Research with chatbots or Google search did not change participants’ trust in institutions, experts, media, or technology, but both modalities increased agreement with progressive views and decreased agreement with conservative views among all participants, regardless of their own political leaning. There was one difference between chatbot research and Google research: the process was 6–10% faster when chatbots were used. The authors caution that they did not test models known to be politically biased; that neither the chatbots nor the search condition reflected today’s AI enhanced, personalized search; and that they studied structured research tasks rather than open-ended political conversation. According to the authors, the findings suggest that for this common use case, chatbots may perform on par with self-directed Google search, with no apparent cost to factual political knowledge, a result relevant to ongoing debates about AI governance and democratic integrity.
PNAS Nexus
Conversational AI increases political knowledge as effectively as self-directed internet search
6-Oct-2026