Generative AI’s Broader Impact Extends to Shifts in Written Language
While concerns about generative AI’s environmental footprint, data privacy risks, and potential cognitive effects dominate public discourse, a subtler linguistic shift is also emerging. Researchers and writers are observing a potential decline in the use of the em-dash—a versatile punctuation mark with a centuries-long history in English prose—in texts AI-assisted or generated by models like ChatGPT.
The em-dash (—) is traditionally used to create emphasis, insert an aside, or denote a sudden break in thought, offering a stylistic alternative to commas, colons, or parentheses. Its use is a staple in long-form journalism, literary fiction, and academic writing for managing complex sentence rhythm.
Preliminary analysis suggests large language models (LLMs) may systematically reformat or avoid the em-dash. This occurs because these models are trained on vast digitized text corpora, where the em-dash is less consistently represented than other punctuation. Furthermore, tokenization processes—where text is broken into manageable pieces for the AI—can treat the em-dash as a less common or more context-dependent token, leading the model to favor more frequently encountered substitutes like the hyphen (-), comma, or period.
This phenomenon, sometimes informally termed the “ChatGPT hyphen,” points to a broader issue: AI writing tools may inadvertently standardize and simplify syntactic structures, potentially eroding nuanced, authorial voice over time. The em-dash, with its specific tonal and rhythmic function, is particularly vulnerable to such homogenization.
Linguists note that language evolution is constant, but the scale and speed of AI’s influence are unprecedented. If AI-generated and AI-revised content becomes a primary source for published material, editorial conventions could shift to align with model output, prioritizing clarity and predictability over stylistic diversity.
The significance lies not in the fate of a single punctuation mark, but in what it represents: the potential for algorithmic training data and processing logic to silently reshape written English. Monitoring this trend is crucial for editors, educators, and writers to preserve grammatical richness and ensure technology augments rather than flattens expressive tools. Future research will need to quantify usage changes and assess whether this reflects a temporary quirk or a permanent stylistic shift driven by pervasive AI assistance.
