Stop AI-Shaming Our Precious, Kindly Em Dashes—Please
Human writers have always used the em dash. In fact, it’s the most human punctuation mark there is.
I stand before you today with violence in my heart. I do not come in peace. I come to obliterate, disparage, and destroy. In this fallen world of ours, there exists certain ideas that must be annihilated before goodness can flourish. I am here to rain holy fire upon one of those ideas, and I am here to do so in the name of a punctuation mark.
The punctuation mark in question is the em dash. The idea—the terrible, mistaken idea—is that the use of em dashes in a piece of writing is a sign that the text was generated by AI. Some people have been saying this on, guess where, the internet. The implication is that human writers should avoid em dashes for fear of being mistaken for chatbots. No. Wrong. I am here to raze this implication to the very ground and salt the earth where it stood.
You're familiar, I hope, with the em dash? I speak of the elegant, elongated hyphen, the gentle friend and ally of all writers, used to set off a chunk of text within a sentence. It looks like this: —. Notice, please, how gracious it is, how welcoming, like an outstretched arm guiding you over a tricky step. It can be used in so many different ways. Perhaps you start a thought and—because thoughts don't always move in perfectly straight lines—find that you need to modify the thought with another thought before you finish it. Perhaps you reach the end of a thought and want to emphasize your conclusion—an em dash can do it. Perhaps you simply wish to mark a pause—a hitch, a breath—in the flow of your thinking. The em dash can be annoying when overused, sure. But when it's used well, it's as flexible and subtle as the turns of thought itself.
Yet for months, the em dash has been the victim of a relentless campaign of online persecution the likes of which no punctuation mark, not even the semicolon, has ever seen. Any chucklehead who disagrees with an online post can, if that post happens to include an em dash, dismiss it with the claim that it was written by a chatbot. It's gotten so bad that people on random subreddits are having to issue PSAs informing their fellow users that em dashes were not invented by ChatGPT. These people are heroes of our time. Normally I'm an enthusiastic advocate of AI-shaming, but I will not stand idly by while you make the em dash a pawn in your dirty game of bot-spotting.
As with many modern plagues, the theory that em dashes belong to the machines has a disputed origin. Some people say it began with a racist-seeming X post. Others say it started on the ChatGPT subreddit. Still others suggest the trend was noticed by people working on AI detection tools, or trying to eliminate bot-spawned slop from forums. Wherever it came from, the theory seems to have started going viral around this past February—objectively the dumbest month, a time when no one should try to have thoughts, much less broadcast them—and soon, em dash discourse was convulsing the internet.
There are two reasons why this discourse must be stopped: The first has to do with the way generative AI works; the second has to do with the fate of the human soul. Let's take them one by one.
More on AI
First, generative AI. I haven't seen any hard evidence that chatbots, in practice, use more em dashes than anyone else. Anecdotally, people say it's hard to get them to stop using em dashes; you can order an AI agent to generate text without dashes, and it might still put dashes in the text it generates. But that only proves that AI chatbots suck, not that human writers use em dashes less frequently.
I can understand why lots of people seized on the idea of a simple AI detection filter. You recognize a vampire because it can't enter your house without an invitation; it would be handy, at a moment when language is suddenly being produced at enormous scale by unthinking machines, to have a similar rubric for text. But looking to em dashes to play this role is like claiming you can identify vampires because they wear black clothes. You're going to confuse a lot of humans for vampires, and you're going to trust a lot of vampires by mistake.
If generative AI does have a predilection for em dashes, though, the reason is simply that many human writers use em dashes. Your chatbot also uses commas, just as human writers do. A chatbot does not have a consciousness. It does not "know" how to write, in any meaningful sense. It doesn't have a style, because style requires thought, preference, and taste. A gen-AI chatbot is trained by scanning gargantuan amounts of text. Based on the patterns it detects in that text, it then assesses the probability that certain words and syntactic constructions will occur in proximity to one another. When you ask a chatbot to write an essay about the architecture of Oslo, it looks for material containing those words, then makes algorithmic guesses as to what other words should appear, and in what order. The reason AI bots generate false results so often—the reason they'll claim buildings are in Oslo that aren't actually in Oslo, or that don't even exist—is that these kinds of probability assessments are crude and imprecise, and include no reliable means of comparing the generated text to reality or truth.
So where did the text on which the bots were trained come from in the first place? Well, until bots started being trained on the output of other bots, all the material they were fed was the work of human writers, and much of it was the work of professionals. Meta, for instance, torrented more than 80 terabytes of copyrighted books. Just straight-up pirated hundreds of thousands, possibly millions, of books. And because em dashes are so useful to writers, books tend to include them. So the bots, programmed to copy the work their creators pirated, use them too.
In other words, it's not accurate to say that the use of em dashes in a text is a sign that the text is AI-generated. It's more accurate to say that the prevalence of em dashes in AI-generated text is a sign of how reliant the AI companies are on the human writers they want to replace.
Now, the second point. Let's talk about what the em dash represents at a moment when it seems as if culture—meaning, for example, the collision of social media, an increasingly authoritarian political discourse, the rapid decline of post-pandemic literacy, the decline of original art, and other related phenomena—has become a massive conspiracy to destroy our ability to think. Grammar and punctuation may seem like boring topics, but you need them if you want to keep that ability alive. Grammar is our system for organizing thoughts into language. Grammar is what lets us put thoughts in relation to one another. The more complex the grammatical structures, the more complex the work they can do with ideas, which is why children's books tend to utilize simpler language than philosophical treatises.
Before the 20th century, a great deal of formal writing instruction concentrated on how to make complex language more readable, elegant, and beautiful. Around a hundred years ago, though, writing instruction started to favor simple, direct sentences and pared-down grammatical structures in the interest of clarity and efficiency rather than complicated thought. Think about the sort of advice you'd get at a college writing lab, for instance, or the sort you'd find in Strunk and White's legendary writing manual, The Elements of Style. Don't use more words than necessary. Keep your sentences short. Use active verbs over passive verbs. Cut down on adverbs and chuck anything that looks ornate straight out the window.
And all this advice is great if you're trying to write an internet column, say, or dash off a quick work email. If you're dealing with more complex ideas, though, the bias of contemporary English toward grammatical minimalism can be somewhat limiting. Without all the shades and nuances of complex sentence forms, you encounter a sort of downward pressure to simplify your ideas—which, again, is often a good thing to do, but not always when you're trying to say something original about the multifarious and contradictory world in which you live.
And that's where the em dash comes to your rescue. This may be a bizarre thing to say, but I think of the em dash as the evolutionary heir of 18th- and 19th-century rhetoric, in the same way that birds are the heirs of dinosaurs. The em dash is a tool that lets writers expand their ideas without making them inaccessible. It's almost always easy to follow in use, and the rules that govern it aren't as ironclad as the rules that govern parentheses or semicolons; you're able to deploy the em dash in more discretionary ways. It lets you break up sentences into parts that rub up against each other, challenge each other, give each other new inflections, or change each other's meanings in ways that resemble the flow of human thought.
It would be a tragedy if writers stopped using em dashes out of fear of sounding like AI, because em dashes are one of the best tools writers have for not sounding robotic in the first place. Their very potential to be irritating is a sign of what makes them so beautiful: Of all the forms of punctuation, the em dash is the one that most rewards tact, judgment, and taste. It has the closest relationship to the way we experience thinking—rushing forward, suddenly swerving, forking into different branches that eventually come together again. If chatbots copy our use of it, they do so for the same reason we need to protect it. It's the most human punctuation there is.
