
(Credit: mark reinstein/Shutterstock)
In A Nutshell
- Researchers analyzed 2.2 million congressional statements and found that lawmakers who specialize in personal attacks receive dramatically more media coverage than policy-focused colleagues.
- Despite the media attention, personal attacks showed no meaningful relationship with fundraising, vote share, legislative output, or personal wealth gains.
- The most combative members of Congress didn’t come from the most politically hostile districts, suggesting they were cultivating a national media audience rather than reflecting constituent demand.
- The study points to media celebrity as an emerging fourth motive for lawmakers, alongside reelection, legislative influence, and good policy, and warns that democratic accountability mechanisms aren’t built to check it.
Voters say they hate it when politicians trade insults instead of ideas. Survey research consistently finds that citizens prefer substantive policy debate over personal attacks. And yet, the personal attacks in Washington keep getting worse. That is the puzzle a new large-scale study set out to crack. If personal attacks don’t seem to win elections, don’t seem to raise money, and turn off voters, why do some politicians keep doing it?
MY LATEST VIDEOS
The answer, according to researchers at Dartmouth College, the University of Pennsylvania, and the University of Notre Dame, appears to have little to do with votes or donors. It’s largely about media attention.
Their study analyzed more than 2.2 million public statements from members of the 118th Congress, the session that ran from January 2023 through January 2025. Floor speeches, press releases, newsletters, and social media posts were all fed into an AI model trained to distinguish genuine policy debate from personal insults. What emerged was a portrait of a specific type of lawmaker the researchers call a “conflict entrepreneur,” a politician who has figured out that personal attacks are one of the fastest roads to cable news coverage. Not to pass legislation. Not to raise money. Just to get seen.
As one retired member of Congress put it: “The most recent additions to Congress don’t care about policy; they care about getting attention.”
How Researchers Measured Congressional Personal Attacks
The team built a dataset linking every public statement from every member of the 118th Congress to records of media appearances, campaign donations, electoral results, and personal finances. Using a large-language model, essentially a sophisticated AI text classifier, they sorted statements into three categories: personal attacks, critical policy debate, and standard policy discussion. A personal attack, as defined in the study published in PNAS Nexus, targets a person or group’s character, honesty, or integrity rather than their policy positions.
Real statements from the paper show what that looks like in practice. Rep. Marjorie Taylor Greene (R-GA) once posted on social media: “Joe Biden is Hitler.” Rep. Jared Moskowitz (D-FL) wrote of Donald Trump: “Your boss [Trump] failed Pictionary when he couldn’t tell the difference between his ex-wife and a woman he assaulted in a dressing room. THE END!” Neither engages with policy. Both went viral.
Two PhD-holding political scientists independently reviewed 500 randomly selected statements to validate the AI’s work. For personal attacks, the model hit 97% accuracy and 98% precision, making it more accurate than commonly used sentiment tools like Vader or TextBlob, which often misclassify political language.
Fame Pays. Votes, Money, and Legislation Don’t Show Clear Gains.
A lawmaker who devoted just 5% of all communications to personal attacks received roughly the same cable news coverage as a colleague who spent 45% of their time on substantive policy debate. The 25 most combative members of Congress collectively received more cable news mentions than the 75 least combative members combined.
Beyond media attention, the numbers go flat in every direction. Conflict entrepreneurs raised no more campaign money than their civil colleagues, with no fundraising relationship turning up for either in-state or out-of-state donations. On Election Day, once researchers accounted for how competitive a district already was, combative lawmakers showed no vote-share advantage at all. They tended to win big not because of their rhetoric but because they came from safe, uncompetitive districts to begin with.
Legislative output told the same story. Conflict-heavy members were less likely to sponsor or cosponsor bills and less likely to land seats on powerful committees. Lawmakers who stuck to policy debate, even sharp policy debate, were significantly more likely to get legislative work done. A review of financial disclosures filed under the Ethics in Government Act found no relationship between combative rhetoric and short-term wealth gains either. The authors leave open the possibility that media celebrity eventually converts into book deals or speaking fees after leaving office, but that falls outside what the data can currently show.
Most Members of Congress Don’t Use Personal Attacks. A Few Do Constantly.
Personal attacks are far rarer in congressional speech than news coverage implies. About 65% of lawmakers used personal insults in less than 1% of their public communications. A small number of outliers drove the entire public perception of how Congress talks.
Republicans made personal attacks at higher rates than Democrats in both chambers, and the House saw more of it than the Senate. Attack rates spiked sharply before elections, climbing from a weekly average of around 0.2% in quieter periods to nearly 5.5% in the final stretch before a general election. Meanwhile, substantive policy debate moved in the opposite direction, dropping from roughly 42% of weekly communications to as low as 7.2% as Election Day approached.
Perhaps most telling, the most combative lawmakers didn’t come from the most politically hostile districts. Researchers estimated partisan animosity across every congressional district using 140,000 survey interviews and found virtually no correlation between a district’s hostility level and how often its representative attacked opponents personally. Some of the loudest members represented relatively moderate voters. Some lawmakers from the most partisan districts barely traded insults at all. These members weren’t giving constituents what they asked for, suggesting some lawmakers may be performing for a national audience of online donors and partisan media, one almost entirely disconnected from the people who actually voted for them.
What Congressional Incivility Means for American Democracy
So why does it keep happening? Political scientists have long said lawmakers chase three things: reelection, influence within Congress, and good policy. What this study documents is the arrival of a fourth: media celebrity. Under the current incentive structure, a lawmaker can build a national following without passing a single bill, winning a competitive race, or out-raising any colleague.
The authors suggest two pressure points worth watching. Parties could limit committee assignments and campaign resources for frequent offenders. Journalists could provide context for inflammatory remarks rather than simply amplifying them.
In the 118th Congress, most lawmakers on both sides still spent far more time on policy than on insults. But the ones who didn’t got most of the coverage, and that may be all the explanation the yelling requires.
Disclaimer: This article is based on a descriptive, observational study. The research identifies associations between lawmakers’ rhetorical styles and outcomes such as media coverage, fundraising, and electoral performance, but does not establish that personal attacks directly cause or prevent any of these outcomes. Findings reflect data from the 118th Congress (2023-2025) and may not apply to other legislative sessions or political contexts.
Paper Notes
Limitations
Because the analysis relied on text-based classification, it almost certainly undercounts the actual volume of personal attacks in Congress. Implicit insults, coded language, sarcasm, visual memes, and indirect attacks through references to a lawmaker’s family all fell outside what the model could reliably detect. Statements blending policy critique with character assessment posed inherent classification challenges, and the authors treat their estimates as a lower bound on the true prevalence of personal attacks. The financial data covers only net worth changes during congressional service, meaning any long-term payoff from media brand-building after leaving office goes unmeasured. Use of a commercial AI model also raises replicability questions, since providers regularly update their systems and future attempts to reproduce the classification may yield different results.
Funding and Disclosures
Support came from the Charles Koch Foundation, the Hewlett Foundation (grant 2023-03308-GRA), the Knight Foundation (grant GR2022-65104), the Carnegie Corporation of New York (grant G-PS-24-61068), and the Templeton World Charity Foundation (grant TWCF-2022-30379). No competing interests were declared.
Publication Details
Marc S. Jacob (Keough School of Global Affairs, University of Notre Dame), Yphtach Lelkes (Annenberg School for Communication, University of Pennsylvania), and Sean J. Westwood (Department of Government, Dartmouth College) authored the study. It was published March 17, 2026, in PNAS Nexus, volume 5, under the title “Entrepreneurs of conflict: A descriptive analysis of when and how political elites use divisive rhetoric.” DOI: https://doi.org/10.1093/pnasnexus/pgag038