ChatGPT as journalistic gatekeeper: GenAI news summaries, sourcing practices, and geographic bias
Year: 2026
Type: article
Abstract: Purpose This study examines ChatGPT as a technological gatekeeper for news and information about the world, scrutinizing its sourcing practices including the extent to which these practices exhibit a geographic bias. Design We employ an audit- and interview-based methodology. We selected ChatGPT for this study because it is the most visited GenAI technology in the world. We selected Bangladesh and the U.S. for this comparative analysis because they represent different standards of press freedom, public-private relationships, and technological infrastructure. Methodology We conducted daily interviews with ChatGPT over a 60-day period using a structured interview protocol. The protocol opens with prompts asking ChatGPT to summarize the latest political news in the U.S. and in Bangladesh. Afterward, open-ended questions ask ChatGPT to explain its decision-making processes, such as why it selected the specific sources for the day’s news summaries as well as broader questions about how it evaluates information sources when creating news summaries. Findings ChatGPT relies most frequently on wire services and Wikipedia when writing news summaries and demonstrates a clear preference for American news sources over Bangladeshi news sources. In addition, ChatGPT said it determines news sources based on five major criteria – relevance, recency, breadth, credibility, and cross-verification – yet identified a sixth criterion – global-local balance – only in reference to its sourcing practices for Bangladeshi news summaries. Implications ChatGPT’s sourcing practices demonstrate a geographic bias such that the model assesses Bangladeshi news sources to be less accessible, less credible, and more in need of regional and international contextualization than American news sources. Originality/value This exploratory study incorporates an original interview-based auditing methodology while providing one of the first empirical studies of ChatGPT’s sourcing practices as a journalistic gatekeeper.
Authors: Md Sazzad Hossain, Brian Ekdale
Institution: University of Iowa
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: diamond
OpenAlex ID: W7204089193