Insults win media attention but not votes, money, or laws - so why do politicians keep using them?

A small subset of US legislators spend a disproportionate amount of their public communications attacking the personal characteristics, motivations, and integrity of other politicians. They do this across floor speeches, press releases, newsletters, and social media posts. And according to a new analysis of 2.2 million public statements from the 118th Congress, the strategy works - but only if the goal is media attention.

Sean J. Westwood and colleagues at Dartmouth College used a large language model to classify congressional communications, breaking them into roughly two-sentence chunks and flagging any chunk that targeted a specific individual or group with criticism of personal characteristics rather than policy positions. The resulting dataset reveals a clear category of politician the authors call conflict entrepreneurs: representatives who lean heavily on personal insults as a communication strategy.

Who are the conflict entrepreneurs?

Republican representatives were more likely to be classified as conflict entrepreneurs than Democratic representatives, but the behavior was not exclusive to either party. Critically, only a small minority of legislators in both parties pursued this strategy to a significant degree. Most members of Congress, regardless of party, communicated primarily about policy.

The distinction matters because it complicates the narrative that American politics is uniformly toxic. The data suggest that divisive rhetoric is concentrated among a relatively small number of practitioners rather than evenly distributed across the legislative body. The perception of pervasive incivility may be driven less by what most politicians do than by the outsized visibility of those who engage in personal attacks.

The visibility payoff - and its limits

Personal attacks do increase media visibility. That finding is consistent with how modern media ecosystems work: conflict generates engagement, engagement drives coverage, and coverage amplifies the attacker's public profile. For a legislator seeking name recognition, the strategy delivers.

But the analysis found no evidence that the enhanced visibility translated into any of the traditional goals of a political career. Conflict entrepreneurs showed no measurable advantage in fundraising. Their electoral margins were no wider. Their legislative productivity was no higher. Their personal wealth did not increase faster than that of their less combative colleagues.

In other words, the insults bought fame but not power - at least not the kinds of power that political scientists have traditionally understood as motivating legislative behavior.

Celebrity as an end in itself

The authors propose an interpretation that, if correct, represents a meaningful shift in how we understand political motivation. For decades, political science has operated under models that assume legislators are primarily motivated by some combination of policy preferences, reelection, and influence within the institution. The conflict entrepreneur pattern fits none of these categories cleanly. Instead, the data suggest that for a subset of legislators, media celebrity has become the primary motivator - not a means to an end, but the end itself.

This framing helps explain a puzzle that has nagged political observers: why do voters who claim to prefer civility and substantive debate consistently re-elect politicians who engage in personal attacks? The answer may be that the attacks are not aimed at persuading voters to support specific policies. They are aimed at generating the media attention that sustains a public persona, and that persona can be maintained independently of legislative accomplishment.

Methodological notes and limitations

The study's use of a large language model for classification is both a strength and a limitation. It allowed the research team to analyze 2.2 million statements at a scale that would be impossible with human coders. But language models can misclassify context-dependent statements, and the two-sentence chunking approach means some attacks may lose nuance when separated from surrounding text.

The study also captures only one term of Congress (the 118th), so it cannot establish whether the conflict entrepreneur phenomenon is growing, stable, or declining over time. Longitudinal analysis would be needed to determine whether media-driven incentives are drawing more legislators toward this strategy or whether it remains the province of a fixed minority.

The finding that media visibility does not translate into fundraising advantages is somewhat surprising, given that small-dollar online donations are often thought to flow toward the most visible and combative political figures. The authors suggest this may reflect the difference between general media attention and the kind of targeted messaging that actually drives donation behavior - a distinction worth further investigation.

The study was conducted at Dartmouth College and involved analysis of floor speeches, press releases, newsletters, and posts on X (formerly Twitter) from members of the 118th US Congress.