Welcome to a higher seminar with the invited guest Cornelia Brantner from Karlstad University. She will talk about her project
How Sourcing Behaviors of Generative AI-powered Search Engines Could Hurt Journalism and Democracy: A Comparative Study of Political News Retrieval in Five Languages
Abstract
In this talk, I will present a study examining news media's role in the context of generative AI-enhanced search engines, focusing on the 2024 Taiwan presidential election. Using Microsoft's Copilot, we analyzed how election news is retrieved across five languages: English, Traditional Chinese, Simplified Chinese, German, and Swedish. Our findings reveal that, despite providing rapid access to synthesized information with an emphasis on transparency, Copilot often generates misinformation and inaccurately attributes sources. A notable pattern emerged in the analysis, showing a heavy reliance on English-language sources—particularly from the UK and US—even when searching in other languages, such as Swedish and Traditional Chinese. This trend raises concerns about the homogenization of information and the marginalization of regional perspectives. The study highlights the tension faced by news media as they balance their role as authoritative sources in democratic societies against the influence of dominant technological infrastructures. We will discuss how AI-driven news retrieval practices impact media diversity, democratic engagement, and the potential repercussions of news outlets opting out of AI crawling. The talk will emphasize the need to enhance transparency, factual correctness, accuracy in source attribution, and accountability in AI-generated content to foster informed citizenship in the digital era.
Biography
https://www.kau.se/en/researchers/cornelia-brantner