Going with the Mainstream: Exploring GPT Representation of Journalistic Culture

Abstract

As generative AI becomes increasingly integrated into journalism, questions about its implications for journalistic practice grow more urgent. Despite the rise of AI systems, news organizations often lack an informed understanding of how these systems operate, potentially undermining journalists’ capacity to exercise agency in AI-assisted news production. This study explores how large language models (LLMs), such as ChatGPT, can reflect journalistic value systems. It does this by prompting the GPT-4o model to respond to survey items from the Worlds of Journalism Study that measure journalistic role perceptions, epistemologies, and ethics. These responses were compared to actual survey data from the US, UK, and Germany. Our findings suggest that GPT-4o’s outputs correspond most strongly to the survey responses of politically centrist, full-time journalists whose employers have a TV background, while showing less alignment with right-leaning, part-time, and non-degree-holding journalists. These propensities were most pronounced in the German dataset. While we do not assert that GPT-4o’s cultural orientation directly translates to its journalistic applications, the results reveal which groups of journalists’ perspectives the LLM is more—or less—likely to reflect. By mapping this alignment, the study offers an empirical point of departure for guiding how journalists can maintain agency in their use of LLMs, and thereby contributes to ongoing discourse about the ethical, epistemological, and institutional dimensions of AI in journalism.