What happens when journalists switch newsrooms? New research from Chicago Booth shows that reporters carry their political leanings with them—shaping news slant in ways we can now measure using AI.
In 2025, Americans can choose to get their daily updates from over 1,000 daily newspaper outlets and more than 10,000 commercial radio stations. These outlets vary widely in which stories they choose to cover and how they choose to cover the same event. Some outlets do so in a way that makes them more favorable to Democrats, while others are more favorable to Republicans—a difference that is often referred to as media slant. Amid this variation and abundance of choice, new research from The University of Chicago Booth School of Business asks: Who determines the political slant of a given article, and do journalists’ identities significantly affect the slant of the news they produce?
Using a Large Language Model (LLM) to measure slant in millions of news articles, researchers found that the identifiable fingerprints of political ideology follow journalists when they change jobs. Led by Chicago Booth economist Jacob Conway, Assistant Professor of Economics, the team turned their attention to the people behind the headlines to better understand if, and to what extent, they bring their political values with them to work.
Conway’s work often considers how individual preferences ripple outward to shape broader market and media behaviors, consequently influencing public discourse. Studying the effects of journalists’ ideologies on slant is imperative because journalists, as a group, do not tend to reflect the broader public: “In many countries, including the United States, the distribution of ideologies among journalists looks quite different than the distribution of ideologies in the general population,” Conway explained. While prior theoretical research has suggested that journalists' preferences could lead to slant and media bias, it hasn’t tested whether this occurs in practice, underscoring the need to understand and empirically quantify this dynamic.
“The people actually writing the articles are the journalists,” Conway noted. “And they potentially have a lot of influence over the slant of news they produce.”
The new study finds that journalists do in fact play a significant role in determining news slant, although outlets ultimately seem to be a more important driver of slant overall.
To better understand the role of journalists, the team turned to artificial intelligence. Conway and his coauthor Levi Boxell focused on how prominent political figures engage with and share news stories as a proxy for identifying whether an article would appeal more to Democrats or Republicans. The team compiled a labeled dataset of over 100,000 articles posted by elected officials on popular social media platforms, which was fed to and processed by the machine learning (ML) model. The model learned to predict the probability that the article would be shared by a Republican vs. Democratic politician, based on the text of the article.
Called RoBERTa, the ML model they used is a type of LLM that learns by analyzing language in context, understanding how the meaning of particular words or phrases might change depending on the words around them. This allows a model to interpret text in a similar way to how a person would, picking up on keywords, tone, and nuanced patterns in phrasing.
“We’re using LLMs in particular, like a BERT-based model—RoBERTa—to try and ‘learn’ from the text of an article whether this article is slanted more in the liberal direction or conservative direction,” Conway explained.
After learning from data on politicians’ engagement with news content, the trained model was able to measure political slant across millions of articles with new levels of precision. The team further validated that, relative to other measures of slant, their model’s predictions most closely aligned with how human readers judged whether articles leaned more Democratic or Republican.
With a base capacity to associate news stories with political slant, the researchers peeled back yet another layer of the media industry to capture what happens when journalists move across outlets. For example, when a journalist transitions from a role at The New York Times to The Wall Street Journal, does their writing style shift to match the tone of the outlet? Or are their individual ideological leanings perceptible and consistent across outlets?
The research found that, on average, journalists retained a significant portion of their original slant even after moving outlets. This finding challenges the idea that respective newsroom cultures are the only determinant of what gets written, instead positing that the individual voices of the journalists also matter.
Conway’s research has contributed to existing literature on the key drivers behind news slant, media bias, and the forces behind public discourse. “Our estimates reject the hypothesis that journalists have zero ideological preferences over the content they produce,” Conway said.
By combining large-scale data with cutting-edge language models, the study was able to help quantify what has been difficult to measure, showing that nearly 16 percent of the observed variation in outlet-level slant can be attributed to journalist preferences alone.
“LLMs have made it possible to quantify things we couldn’t before,” Conway said. He anticipates that the use of LLMs will continue to be a vital tool in research analysis and pattern discovery in political communication. But these findings also raise broader questions—not just about individual reporters, but about the evolving role of AI in media analysis.
How might journalists’ ideologies evolve over the course of their careers? Are there group dynamics within newsrooms that shift how writers express their views? And how might editorial policies amplify or dampen individual slant? Questions like these pave the way for future investigation, especially as LLMs become more capable of tracing subtle shifts and patterns across massive volumes of text. “LLMs,” Conway noted, “are just beginning to show what’s possible.
Using data-driven inquiry to apply artificial intelligence to real-world challenges is central to the mission of the Center for Applied Artificial Intelligence. Deeper, machine-powered analysis has the potential to reveal the hidden dynamics behind the headlines and to support critical engagement with the systems and platforms that shape public thought and discourse.
The next time readers scroll through an article, it’s worth considering how much of the story is shaped by the journalist’s views versus the facts themselves. As AI reveals the nuances of media bias, it’s clear the future of news is not just about what’s being reported—it’s about who’s telling the story.
Read the full research publication here.
