Readers rarely notice punctuation — until it gets in the way. The em dash links clauses that could stand on their own, replacing punctuation that might break up the paragraph in a more traditional way. It emphasizes what follows after the mark — seeming abrupt and dramatic in some cases. While flexible and commonly used in human writing, it has also been the subject of scrutiny recently due to its repeated usage in generative AI writing. Its usage isn’t wrong — but many unfortunately consider it a telltale sign of AI usage, creating a stigma around its use.
With their versatility, em dashes can be used as alternatives for other punctuation types such as commas, colons and parentheses, but their ease of use is responsible for their misuse as well. When used properly, they either connect two independent clauses together (like a colon), or connect a main clause and a subordinate clause (more similarly to how a semicolon or comma functions). They are even used to introduce quotes in some cases. Despite their usefulness, they’ve had a rocky history even before AI.
From Typeset to Text
Em dashes are different from other dashes due to their width. They derive their name from typography and are traditionally the width of a capital “M” in a given typeface, hence the name. In early typesetting, the em dash was used to create interruptions or long pauses, which helped save space by bridging sentiments. While they were popular in the early days of printing, they didn’t have a designated em dash key when typewriters came around. To designate an em dash, writers had to use double hyphens (“–”), leading to hyphens being preferred over em dashes. As a result, the em dash gradually became less common, with usage being limited mostly to formal print until the rise of word processors and digital typography. Since then, though not a standard key on the keyboard, it has become significantly easier to use.
Soon after the rise of digital typography came the internet, which hosted writing of all styles from people worldwide. For them, em dash usage offered a quick, versatile punctuation mark for injecting emphasis without the rules of a semicolon or separation of parentheses. Word processors and the ability to set shortcuts also made the em dash more approachable than ever. Despite this, not all writers choose to adopt their usage.
Generative AI models such as OpenAI’s GPT are trained on publicly available and licensed data. This can result in whatever web content was publicly accessible at the time of training influencing the actions of the model. AI is trained to mimic human writing, especially since it is trained on it. Humans pause mid-sentence when we talk, we go off on tangents and we don’t always keep a consistent line of focus. While that may seem unrelated, many of these traits carry into human writing as well, such as rhythm, cadence and writing style. AI models don’t “hear” what they’re writing and don’t know what a pause feels like — they’ve just seen what one typically looks like. While trained on massive amounts of human-written text, the repeated use of the em dash stems from its use across various styles and genres of writing. Because of its versatility, it became a multi-purpose tool for AI models; it offers a flexible pause, sidestep or substitute for actual stylistic judgement. Without an internal rhythm, the model leans on what’s most common and versatile without being wrong, leading to repeated em dash usage.
For many, the overuse can be seen as robotic or uncanny, a mere imitation of human writing that large language models attempt to pass off as neutral and human-sounding. The em dash draws attention to itself when repeated, particularly in an informal tone or in other contexts where they aren’t commonly used. While readers may not notice it initially, as they read through, they might feel something is off. Its use is not grammatically wrong most of the time, but we pick up on the patterns of automation, especially for skeptics of AI-generated writing. It is not a surefire sign of AI usage, but a mixed signal that writing might be machine-generated.
While the em dash began as a space-saving tool that also happened to add dramatic effect to writing, it now serves as a mirror of how writing is evolving due to the influence of generative AI. Current overuse is not the fault of writers nor machines, but the circumstances in which the models are trained and the result of a massive push for artificial intelligence in everyday life. If anything, it serves to show how useful the em dash can be for writers, capturing both the directness of modern writing and the unpredictability of machine-learned prose.
