Zhang, Xiaoke, Myunghwan Lee, Mi Zhou, Gene Moo Lee “News Speed vs. Quality: Investigating Large Language Models’ Impact on Modern Journalism,” Work-in-progress.

• Presentations: UBC (2024), DS (2024), CIST (2024)

With the advancement of generative artificial intelligence (AI), news outlets are increasingly incorporating large language models (LLMs) into their workflow to increase news productivity and quality. Utilizing a unique empirical setting where two major news organizations in South Korea introduced LLM-based news writing assistants, this study examines how LLM assistance affects news production and consumption. We first developed a novel framework using GPT-4o to extract information sources from news articles. We then constructed a unique dataset of 571 LLM-assisted news articles and 3,489 competing human-generated articles covering the same events. Using the DiNardo-Fortin-Lemieux reweighting method to ensure comparability between the LLM-assisted and human-generated news, our empirical analysis reveals that LLM assistance significantly increases news publication speed but reduces the diversity of information sources in news articles. Furthermore, LLM-assisted news is associated with decreased reader consumption, a trend exacerbated by reduced source diversity even with faster publication speed. Our findings contribute to the broader literature on generative AI’s role in professional content creation.