Summary: Researchers at Virginia Tech applied artificial intelligence and natural language processing to a decade of broadcast transcripts and social media activity from CNN and Fox News. Their analysis reveals a measurable increase in partisan and inflammatory language over time and shows how broadcast language shapes online public debate.
The study demonstrates that the words broadcast on cable news influence conversations on social platforms, contributing to polarized echo chambers. It also finds that audiences of different networks often interpret the same words in very different ways, reinforcing preexisting viewpoints and widening civic divisions.
Key Findings:
- The researchers analyzed nearly 300 billion words spoken on CNN and Fox News broadcasts and nearly 133,000 tweets connected to six politically contentious topics between 2010 and 2020.
- Audiences of the two networks not only hold different political views but also attach different meanings to the same keywords. For example, immigration discussions on Fox News tended to emphasize terms like “illegal,” “enforcement,” and “order,” while CNN coverage more often used language such as “family,” “parents,” “children,” and “communities.”
- Although the empirical evidence is robust, persuading broadcasters to change incentive-driven business models that prioritize ratings over civic deliberation remains a major challenge.
Source: Virginia Tech
Overview: National coverage on the two largest cable news networks, CNN and Fox News, not only mirrors growing political polarization in the United States but, according to this Virginia Tech study, has also become more partisan and inflammatory over time. The research links changes in broadcast language to shifts in online public discourse, suggesting television language can exacerbate division in the public square.
Collaborative insights
Eugenia Rho, an assistant professor in the Department of Computer Science with training in political science, led the computational analysis. Her work combines AI techniques with social science questions to better understand how institutions and media shape public conversation.
“When language is repeatedly broadcast by influential actors, it changes the way the public discusses social issues,” Rho said. “Rigorous analysis of very large data sets opens new avenues to understand media effects.”
Mike Horning, a journalist-turned-academic and associate professor in the School of Communication, contributed expertise on misinformation and civic participation. He highlighted how computational methods now make it possible to analyze media at a scale previously unimaginable.
“In the past, studying media bias might rely on a few hundred articles. Now we can process terabytes of data and ask far more consequential questions about media and democracy,” Horning said.
Big data, nuanced knowledge
The team used natural language processing to assess whether broadcast news contained partisan and inflammatory content, how that partisanship evolved from 2010 to 2020, and whether broadcast language shaped debates among viewers on social media. Two primary data sets were analyzed:
- Closed-caption transcripts from CNN and Fox News programming, broadcast continuously between Jan. 1, 2010, and Dec. 31, 2020, provided by public archives and media analysis projects.
- Tweets from 2010–2020 authored by users who followed both @CNN and @FoxNews, mentioned or replied to either account, and included keywords tied to six contentious topics: racism; Black Lives Matter (BLM); policing; immigration; climate change; and health care.
By comparing the language patterns in broadcast transcripts with the language used by audiences online, the researchers traced how certain words and phrases spread from TV to social platforms and how audiences adopted different framings depending on their preferred network.
Democracy in the balance
The study underscores that television remains a dominant news source for Americans—viewers access TV news far more often than print or online outlets—and that people often choose news sources that align with their political identities. The researchers show that broadcast language predicts how viewers discuss national issues on social media and that those online conversations can, in turn, reflect broadcast framing.
Importantly, the analysis found that audiences interpret the same terms differently. For example, coverage of immigration on Fox News commonly featured enforcement-oriented language such as “illegal” or “order,” while CNN coverage more frequently emphasized human-centered words like “parents” and “children.” In discussions of race, CNN frequently referenced “protests,” whereas Fox News often associated the topic with “crime.”
Within two to three months, Twitter audiences mirrored the language patterns of the broadcast news they favored, reinforcing separate information environments and deepening echo chambers. These divergent framings make it harder for citizens to find shared language for public debate.
“Words have immense power and tangible effects on civic life,” Rho said. “When major networks portray the same topics in almost entirely different terms, audiences end up inhabiting distinct realities. That divergence makes constructive public conversation difficult.”
Economic pressures help explain why this pattern persists. As cable news competes for dwindling viewership and online attention, sensational or polarizing language can attract audiences and boost ratings, creating incentives to favor market-driven content over deliberative civic coverage.
“Declining viewership and fierce competition incentivize outlets to stand out,” Horning said. “That often means becoming more provocative to cut through the noise, with consequences for democratic discourse.”
Real-life impact
The central takeaway is that partisan broadcast news is a measurable contributor to political polarization among the electorate. The research provides a decade-long, data-rich basis for that claim, but changing newsroom incentives or programming strategies will be difficult.
Rho emphasized the civic importance of the findings: “Demonstrating these patterns at scale should spark conversations about what kinds of media practices support collective problem-solving. If we cannot agree on how to talk about core issues, democratic progress is hampered.”
This analysis can also help viewers make more informed choices about their media diets by revealing how language shapes perception and debate.
About this political psychology research news
Author: Margaret Ashburn
Source: Virginia Tech
Contact: Margaret Ashburn – Virginia Tech
Image: The image is credited to Neuroscience News
Original Research: “Same Words, Different Meanings: Semantic Polarization in Broadcast Media Language Forecasts Polarity in Online Public Discourse” by Eugenia Rho et al., Proceedings of the Seventeenth International AAAI Conference on Web and Social Media (open access).