When it comes to distinguishing between human-written and A.I.-generated text, you might expect to look for complex patterns. Yet, earlier this year, the internet became fixated on something surprisingly simple: the em dash. ChatGPT, it seemed, couldn’t get enough of them—using them constantly, even when explicitly told not to.
As this observation spread, a peculiar consensus emerged: humans, apparently, don’t use dashes. Tech forums buzzed with users calling them a ‘GPT-ism’ and ‘robotic artifacts’ that were ‘out of sync with modern communication.’ One user on an OpenAI forum even lamented that the excessive dashes made it difficult to integrate ChatGPT into customer service without human users noticing. It seemed countless individuals were mystifyingly convinced that no flesh-and-blood person had any real use for this punctuation mark, and that anyone who dared to use one would henceforth be mistaken for a machine.
Naturally, those of us who appreciate the humble dash were appalled. I, for one, am a former proofreader who could passionately extoll the virtues of the narrower en dash. While I might concede that my ‘dash-happy’ lifestyle is somewhat atypical, I was genuinely surprised to hear the very existence of the dash questioned. The dash is a time-honored, entirely normal tool for crafting sentences! Just look at Dickens, Dickinson, Nietzsche, Stephen King novels, or even this very magazine—they’re all liberally sprinkled with dashes. In fact, part of their enduring popularity stems from their ability to make writing feel more casually human, closer to natural speech, than the more formal colons, semicolons, and parentheses. Humans don’t typically think or speak in perfectly structured sentences; we think and speak in interconnected thoughts, which often interrupt, introduce, and complicate each other in an intricate dance that forms larger, more complex ideas. (Sometimes it doesn’t, of course: J.D. Salinger’s dialogue, with its abundant dashes, beautifully illustrates just how often our thoughts are cut short or left hanging.) This, ultimately, is what punctuation is for.
The most telling A.I. signal isn’t about the dash itself, but its presentation. ChatGPT formats its em dashes in the traditional style of a printed book—a stroke the width of the letter ‘M,’ with no spaces on either side. The average computer user, however, doesn’t typically type this way. Many might not even know the specific keystrokes required to produce this character (or its proper name; some online discussions referred to it as a ‘ChatGPT hyphen’). Instead, most users simply insert a hyphen (-) or two (–), which some software might then auto-correct to the often-overlooked en dash (–). Crucially, the average user tends to put spaces around their dashes, just as most online publications do, which helps text flow more smoothly across lines.
Oceans of communication that used to be handled by speech are now left to lone individuals typing into the internet.
Despite these nuances, the debate continued to circle back to the dash itself. People discussed it as if it were some mystical, archaic symbol that no self-respecting human would ever consider using. “Nobody uses the em dash in their emails or text messages,” one commenter insisted. “This punctuation is irrelevant to everyday use-cases.”
I’m not writing this to mount a grand defense of dashes. Instead, I want to propose that the phrase ‘everyday use-cases’ hints at a truly monumental shift in our fundamental understanding of what writing is.
For a significant period of recent history, most written material people regularly encountered—anything beyond simple signs and menus—was what we might call ‘writing-writing.’ This was text that someone had consciously composed, perhaps revised or edited, and potentially even had professionally printed. This form of communication was distinctly different from our daily interactions with peers; we primarily spoke to them. Even with the advent of the internet, this basic psychological arrangement largely persisted.
But now, that distinction has all but vanished. ‘Emails or text messages,’ social media posts and chats, DMs and comments, even DoorDash drivers informing you that a restaurant is out of coleslaw: vast oceans of communication previously handled through speech are now entrusted to individuals typing into the digital void. Even if you consider yourself a dedicated reader, you likely spend more of your time engaging in these spontaneous, on-the-fly typings, because that has become the everyday reality of writing.
This everyday language is still wonderful—so delightfully expressive that it’s even developed its own equivalent of the sarcastic, mocking tone we use in spoken language. (You know, tYpInG LiKe tHiS.) But ‘writing-writing’ is a different beast, isn’t it? At its best, it taps into a different intellectual register: perhaps less immediate and visceral, but often more refined and deliberate, more artfully constructed, meticulously choreographing the flow of thought with greater precision, depth, and, yes, punctuation.
Large language models are trained on colossal quantities of human-generated prose, encompassing far more historical printed matter than most of us will ever encounter. We instruct these machines to emulate our writing, but we don’t always specify—and may not even fully grasp—that our current definition of ‘writing’ now heavily includes the almost oral communication we constantly toss back and forth across our screens. When we then scrutinize the A.I.’s output, we spot the telltale signs of books and magazines, and ironically, we perceive these traditional artifacts as faintly robotic. These machines are simply reflecting our own traditions back at us. What we may not yet comprehend is that we are, in turn, drifting towards entirely new ones.
Illustration by Outlanders Design