A new linguistic analysis published in PLOS One suggests that Donald Trump and Kamala Harris employed distinct rhetorical strategies during the U.S. presidential debate—but they also shared several unexpected linguistic similarities. The findings indicate that some stylistic differences commonly associated with Republican and Democratic rhetoric may have blurred during the high-stakes debate.
Political debates provide a unique opportunity to analyze how candidates craft their language in real time, under pressure, while addressing voters from across the political spectrum. The researchers behind the study aimed to determine whether Trump and Harris’ rhetorical styles aligned with broader patterns typically associated with their respective parties.
Previous work suggests that Democrats tend to emphasize inclusivity, empathy, and collective responsibility, while Republicans focus on individual authority, traditional values, and strength. The new study sought to examine whether such tendencies were reflected in the candidates’ language during the 2024 debate.
“My research group is funded by an EU grant for the ABSTRACTION Project, where we investigate how language and word choice in particular shapes abstract thinking and social interaction,” explained corresponding author Marianna Bolognesi, a professor of linguistics at the University of Bologna and principal investigator of the ABSTRACTION Project.
“The 2024 Trump–Harris debate offered a valuable case study to test whether the ‘red’ and ‘blue’ language styles identified in previous research would persist when both candidates were confronted with the same questions under identical constraints. We expected to find clear differences, but instead observed far fewer than anticipated, suggesting that both candidates adjusted their language to the context of the debate.”
For the study, Bolognesi and her co-author Philipp Wicke analyzed full transcripts of the September 2024 debate between Trump and Harris, broadcast on ABC News. The researchers used a mixed-methods approach combining large language models, lexical databases, and human annotation. They assessed over 100 individual responses from Trump and Harris, applying both manual and automated tools to identify linguistic patterns.
Among the methods used were GPT-4o for figurative framing detection, SiEBERT for sentiment polarity analysis, and DeBERTa for political ideology classification. Word specificity was assessed using the WordNet hierarchy, while concreteness was evaluated using human-rated norms. The researchers also manually counted pronoun usage and tracked the number of direct references each candidate made to their opponent.
One of the most striking contrasts between the two candidates lay in their use of figurative frames. Trump frequently described the nation as being under siege, using metaphors of collapse and invasion. He portrayed immigration as a “flood” or “invasion,” the economy as being “sold down the tubes,” and societal decay as a “bloodbath” or “crime through the roof.” This framing leaned on hyperbole and catastrophe, aiming to evoke fear, urgency, and a need for drastic change.
Harris, on the other hand, used frames focused on resilience, rebuilding, and unity. She described economic recovery as “lifting up the middle class” and referred to American democracy as something to be “defended” and “restored.” Her use of metaphor tended to focus on collaboration and moving forward, such as “charting a new way forward” or “cleaning up Donald Trump’s mess.”
Harris used figurative framing more frequently than Trump. On average, she employed nearly three frames per response, while Trump used about one and a half. These frames were validated through a combination of large language model analysis and manual annotation, with high agreement between human raters.
Contrary to expectations, there were no meaningful differences in lexical specificity between the candidates. Both Trump and Harris used language with similar levels of detail, based on the depth of words within the WordNet semantic hierarchy. This finding challenges the assumption that Democrats tend to be more policy-dense and Republicans more slogan-driven in debates. It may reflect the fact that both candidates were responding to the same questions and attempting to appeal to a national audience.
“Most of the findings were surprising, but one unexpected result was that Trump and Harris did not differ significantly in the specificity of their answers, both blended general claims with specific examples,” Bolognesi told PsyPost.
Similarly, both candidates predominantly used abstract language. While Harris’ word choices were marginally more concrete, both fell near the abstract end of the concreteness scale. The authors suggest this may reflect a broader strategic aim: abstract language can invite broader interpretation and engagement, while more concrete language may narrow a candidate’s message.
“The more interesting divergence emerged in word abstractness, that is, in their use of terms referring to intangible entities: Harris remained fairly consistent in relying on abstract terminology, while Trump shifted more sharply between concrete terms such as jobs and borders and sweeping abstractions like American greatness,” Bolognesi said. “This was unexpected, since prior research led us to anticipate Harris would rely more on abstract but specific terms, and Trump on concrete but generic language.”
While both candidates relied heavily on emotionally charged language, the study found no statistically significant difference in the overall emotional polarity of their responses. Both Trump and Harris tended toward negative sentiment, a pattern that aligns with prior research showing that modern political communication often uses fear and outrage to galvanize audiences.
However, Harris was found to use more subjective language than Trump. Her responses included more expressions of personal judgment and moral perspective, while Trump’s statements were more likely to be framed as statements of fact, even when they involved speculation or exaggeration. This subjectivity was measured using two independent models, both of which yielded consistent results.
“We also found no significant difference in the emotional valence of the words they used: both relied heavily on terms with slightly negative connotations, contrary to our expectation that Trump would be more negative and Harris more positive,” Bolognesi said. “Another striking finding was that Harris used more subjective language, framing issues in terms of empathy, moral judgment, and personal perspective, while Trump’s rhetoric appeared more objective, often presented as statements of fact.”
The researchers uncovered marked differences in how each candidate addressed their opponent and constructed social identity. Trump did not refer to Harris by name even once during the debate, while Harris mentioned Trump by name 70 times—more than half of her responses included direct references to him. This contrast likely reflects differing rhetorical strategies: Harris leaned into confrontation and accountability, while Trump avoided direct acknowledgment of his opponent, possibly to minimize her perceived legitimacy.
In terms of pronouns, Trump used first-person singular forms like “I” and “me” 245 times, while inclusive pronouns like “we” and “us” appeared only 133 times. Harris showed a more balanced ratio, using singular pronouns 144 times and inclusive ones 136 times. This supports the hypothesis that Republican rhetoric tends to emphasize individual leadership, while Democratic discourse emphasizes collective responsibility.
Using a political ideology classifier, the researchers assessed whether each candidate’s statements aligned more with Democratic or Republican ideals. As expected, the vast majority of Trump’s responses were labeled Republican, while Harris’ leaned Democratic.
“Word choice is never random: it shapes how issues are framed, and determines which audiences feel aligned and which feel alienated,” Bolognesi told PsyPost. “As expected, Harris tended to emphasize empowerment and recovery with positive figurative frames and metaphorical language (“we will move forward”), while Trump leaned on crisis narratives (‘our country is collapsing’), a pattern confirmed by our statistical analyses. We also found that Trump relied more on singular pronouns (‘I built’), whereas Harris balanced singular and plural forms (‘we will fight together’). Finally, Trump avoided mentioning Harris by name, while Harris frequently invoked Trump directly, another telling difference in rhetorical style.”
While the study provides a comprehensive look at how language shaped the 2024 debate, some limitations are worth noting. The use of automated tools, while powerful, may miss nuances in meaning or context. The sample also reflects just one debate, meaning the findings might not generalize to other events or forms of political speech.
“One key point we highlight in the interpretation of our findings is that this was a presidential debate addressed to the entire electorate, not a partisan rally,” Bolognesi noted. “Because the audience was broader and more heterogeneous, both candidates likely adjusted their linguistic strategies, which helps explain why some of the differences we expected based on previous research were less pronounced. This also shows that both candidates possess a flexible vocabulary, to adapt to different communicative contexts.”
Future research may investigate how these linguistic patterns shift across multiple debates or differ from spontaneous remarks made during town halls or campaign rallies. The researchers are also examining non-political contexts to determine whether similar rhetorical strategies appear outside the political arena.
“Political language is just one of the discourse genres we are examining,” Bolognesi explained. “Within the ERC ABSTRACTION project, we are also studying other domains—for example, everyday conversations, where we’ve found that abstract words tend to spark curiosity and keep dialogue going, while concrete words often close the exchange. We are also analyzing how the same scientific topic is explained differently depending on the audience’s expertise (children, adults, experts, non-experts), and how these choices shape understanding. Another strand of our work focuses on the development of language-mediated abstraction skills in children, exploring how word use influences their ability to think and reason abstractly.
“One of our key standpoints is that effective communication requires a repertoire of words across different levels of abstraction, allowing speakers to adapt to different audiences—whether experts, non-experts, children, or voters,” Bolognesi added. “In the debate, both Trump and Harris showed this skill, aligning more closely with each other than with the partisan styles typical of campaign rallies. To foster this ability in children and adults, in both English and Italian, we developed Word Ladders, a research-based linguistic game that trains users to move flexibly between concrete and abstract words. The app is free, ad-free, and available here: https://www.abstractionproject.eu/playthegame.”
The study, “Red and blue language: Word choices in the Trump and Harris 2024 presidential debate,” was published June 3, 2025.
