log in  |  register  |  feedback?  |  help  |  web accessibility
Downstream Effects of LLM Summarization and Paraphrasing on Human Beliefs and Information Behaviors
Wednesday, December 9, 2026, 11:00 am-12:00 pm
  • You are subscribed to this talk through .
  • You are watching this talk through .
  • You are subscribed to this talk. (unsubscribe, watch)
  • You are watching this talk. (unwatch, subscribe)
  • You are not subscribed to this talk. (watch, subscribe)
Abstract

Large language models (LLMs) are increasingly becoming the interface through which people encounter information online, from LLM-summarized search results (e.g., Google AI Overviews) to LLM-paraphrased persuasive text for information campaigns. This talk presents a line of work examining how LLM-generated content shapes users’ beliefs and information behaviors. The first study, a preregistered randomized controlled experiment (N = 2,004), shows that exposure to AI-generated one-sided summaries shifts users' attitudes, behavioral intentions, and policy support toward the summary's framing across four contested health topics. The second study, two preregistered randomized controlled experiments (N = 2,797) across four contested policy issues, finds that AI search summaries reduce engagement with the search results beneath them, and in some cases influence users' attitudes. The third preregistered study (N = 1,200) demonstrates the persuasive potential of AI-paraphrased information at scale, showing how LLMs can lexically diversify repetitive propaganda, being effective at increasing perceived consensus around the campaign’s broader false narrative, while evading current AI-text detectors. Together, these studies suggest that LLMs are changing not only how people find information but also what they come to believe, with real consequences for public health, politics, and beyond.

Bio

Dr. Yiwei Xu is Assistant Professor in the College of Information at the University of Maryland, College Park, also affiliated with the Institute for Trustworthy AI in Law & Society (TRAILS) and the Artificial Intelligence Interdisciplinary Institute at Maryland (AIM). She received her Ph.D. in Communication from Cornell University and completed postdoctoral training at the University of Washington Information School. Her research sits at the intersection of human-AI interaction, information behavior, and health; combining experiments, surveys, content analyses, and computational methods. Her work appears in venues such as PNAS Nexus, Health Communication, Political Communication, Milbank Quarterly, JMIR AI, AAAI ICWSM, EMNLP, etc. Her research has been supported by grants from the National Science Foundation, and she has received prestigious paper awards from the International Communication Association.

This talk is organized by Sarah Wiegreffe