The em-dash em-beds in Congress: A population-level rise in em-dash frequency in U.S. congressional press releases at the dawn of the large-language-model era, 2021-2025

Research from arXiv suggests a measurable increase in the use of unspaced em-dashes in U.S. congressional press releases between 2021 and 2025. This stylistic shift is identified as a potential indicator of large language model (LLM) assistance in text generation, given that this specific em-dash usage deviates from standard U.S. press style guidelines, such as AP style, but is common in typeset English prose often associated with LLM outputs. The study analyzed over 146,000 press releases from House and Senate offices, observing a consistent density of 0.10-0.12 unspaced em-dashes per 1,000 characters.

Why it matters

The potential for stylistic markers to identify AI-generated content in official communications has significant implications for authenticity, transparency, and trust in public discourse. Understanding the prevalence and nature of AI assistance in governmental communications is crucial for maintaining public confidence and ensuring accountability in the information disseminated by public officials.

Key insights

• A population-level rise in the frequency of unspaced em-dashes (word---word) has been observed in U.S. congressional press releases from 2021-2025.
• This specific em-dash usage is considered a stylistic trace potentially indicative of large language model (LLM) assistance.
• The observed usage deviates from standard U.S. press writing conventions, such as AP style, which specifies spaced dashes.
• Analysis covered 146,239 press releases from 480 House and Senate offices.
• The density of unspaced prose-form em-dashes remained within 0.10-0.12 per 1,000 characters of cleaned text during the study period.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.05889