Zhang, Xiaoke, Myunghwan Lee, Mi Zhou, Gene Moo Lee. “Large Language Models in the Institutional Press: Investigating the Effects on News Production and Consumption,” R&R, MIS Quarterly.

• Presentations: UBC (2024), DS (2024), CIST (2024), BIGS (2024), JUSWIS (2025)
• Industry partner: Muhayu

The rapid advancements of Large Language Models (LLMs) have introduced new opportunities and challenges for the institutional press. Utilizing a mixed-method approach, this paper combines two qualitative studies and two sets of large-scale quantitative studies to theorize and empirically examine how LLM assistance affects news production and consumption. We begin with open-ended surveys and interviews with 12 journalists to identify three key constructs central to the journalistic value–source quality, publication promptness, and reader engagement – and formulate our research questions. To empirically examine these dimensions, we compile a comprehensive dataset of 2,060,894 news articles sampled from 111 major South Korean media outlets. We collaborate with industry experts to fine-tune a Korean-language LLM detector to identify undisclosed LLM usage in the news corpus and leverage GPT-4.1 to label information sources in each article. Our event-level analysis reveals that while LLM assistance expedites news publication, it is associated with a reduction in both the number and quality of sources, as well as a decline in reader engagement. To further investigate the impact of LLM adoption on journalists’ long-term information sourcing behaviors, we conduct a journalist-level analysis using staggered Difference-in-Differences. Results reveal that journalists reduce the use of primary, unaffiliated, and contextual sources after LLM adoptions, and alarmingly, these negative effects are enlarging over time. Drawing on the information foraging theory and a second-wave of qualitative study with 32 journalists, we explore the underlying mechanisms and posit that the negative effects of LLM adoption on source quality are driven by a combination of LLM limitations and long-term shifts in journalists’ behaviors. We conclude by proposing actionable guidelines for the institutional press on combining technical solutions with organizational policies to mitigate the negative effects and facilitate the responsible integration of LLMs. Our findings contribute to the growing literature on the digital transformation of journalism.