Em-dash frequency in U.S. congressional press releases doubles in 2025, study finds

A recent study, available on arXiv (2608.05889), reveals a notable uptick in the frequency of unspaced em-dashes in U.S. congressional press releases beginning in 2025, aligning with the emergence of large language models (LLMs). Analyzing 146,239 press releases from 480 House and Senate offices between 2021 and 2025, the research found that the rate of unspaced em-dashes per 1,000 characters remained consistent at 0.10-0.12 from 2021 to 2024, before surging to 0.217 in 2025—over twice the four-year average. The proportion of releases featuring these em-dashes also increased from around 13% to 24.8%. Conducted with Poisson/negative-binomial models and preregistered with the Open Science Framework (OSF: 10.17605/OSF.IO/U5NEY), the study suggests that LLMs may influence political communication styles, as unspaced em-dashes are common in typeset English but rare in U.S. press writing, which follows AP style guidelines. The authors propose that this observable shift could indicate LLM usage in formal communications.