For most of my life I didn't know the em dash existed. Not really. I'd been using some version of it (probably a hyphen doing several jobs at once) my whole writing life without a single thought, and if you'd stopped me on the street in 2019 and asked me to draw one, I would have given you a slightly-too-long hyphen and moved on with my day. Now, in 2026, it might be the single most discussed punctuation mark in the English language, and the discussion is not "isn't it elegant" but "please, for the love of god, stop using it." That's a strange kind of fame. We spent roughly three hundred years not particularly caring what the em dash was for, and about eighteen months deciding we hate it.
I have a slightly embarrassing credential to bring to this conversation. I did a Masters in graphic design with a genuinely nerdy focus on typography: type anatomy, kerning, tracking, the works. I can tell you why a lowercase "a" in one typeface reads as friendlier than the same letter in another. I spent hours nudging letterforms half a point closer together because the gap looked wrong to my eye before I could explain why it did. And in all of that study, the em dash got, at best, a footnote. I knew the en dash was the width of a lowercase "n" and the em dash was the width of a capital "M," and that was roughly the extent of my formal education on the subject. Nobody sat me down and explained when to reach for one over a comma or a colon. It was a tool in the box I was vaguely aware of and essentially never opened.
So it's a bit rich that this modest little mark, one I could barely define a few years ago, is now something I actively scan my own writing for and delete on sight, purely because I don't want to be mistaken for a chatbot. I didn't learn to love the em dash and then fall out of love with it. I went straight from indifference to suspicion, and I suspect most of you did too.
Where the thing actually came from
Turns out the em dash is older than any of this drama by a few centuries. Its name comes straight from the printing trade: in the days of metal type, a dash cut to the width of a capital M was called, sensibly enough, an em dash, and the shorter one cut to the width of an N became the en dash. Printers were working with these measurements as far back as the fifteenth century, not for any grammatical reason, but because typesetters needed a reliable unit for spacing type consistently. The rules came later. The tool came first.
By the eighteenth century, printers and writers had turned that practical measuring stick into a genuine stylistic device: a way to interrupt a sentence, insert a stray thought, or let something trail off with a bit of drama. And nobody used it with more commitment than Emily Dickinson, whose poems are so thick with dashes that literary scholars half joke the "em" could stand for her name. Her editors, when they first published her work in 1890, stripped most of those dashes out and replaced them with commas and periods, because they thought her original punctuation looked undisciplined. It took decades before anyone put them back.
I find that detail almost too good. A hundred and thirty years ago, editors were removing dashes to make writing look more properly, humanly composed. Today we're doing exactly the same thing, for exactly the opposite reason: deleting dashes to prove a human wrote it at all.
So who is actually in charge of any of this
Here's a question I hadn't seriously considered before starting this piece: who decided any of this? Who is the supreme court of English punctuation, quietly ruling on whether a semicolon or an em dash is the correct move in a given sentence?
Nobody, is the honest answer. English has no equivalent of the Académie française, no government office that issues punctuation law. What we have instead is a handful of competing style guides (the Chicago Manual of Style, the AP Stylebook, Oxford, Fowler's, and half a dozen others) that each offer strong, occasionally contradictory opinions, plus a general sense of consensus that builds over time from what enough people agree looks right. Grammar, in English, is closer to fashion than to law. It's set by usage, reinforced by teachers and editors and style guides who all have slightly different tastes, and it drifts. The Chicago Manual, for what it's worth, calls the em dash "the chameleon of punctuation marks," capable of standing in for a comma, colon, semicolon, or set of parentheses, which is either its greatest strength or the reason it's completely out of control, depending on your mood that day.
So no one is the boss. There's no supreme authority to appeal to when you want to know, definitively, whether that dash was correct. There's just taste, convention, and a very large group of increasingly opinionated readers.
Why does the robot love it so much
This is the part I actually went looking for an answer to, because "AI just likes it" isn't a real explanation. It turns out there is a real one, and it's more interesting than I expected.
Large language models don't write the way we do. They predict, one small chunk of text at a time, which piece is most likely to come next given everything before it: the world's most sophisticated version of the autocomplete on your phone. When these models were trained, engineers scored huge amounts of writing for quality. A celebrated novelist's prose might rate a ten out of ten, a professor's essay an eight, an average blog post a three. And it turns out the writers producing that top tier of text, the novelists, the essayists, the long-form journalists, use the em dash constantly. It's a favourite tool of exactly the kind of writing that gets marked as excellent. So when the model reaches for its highest quality options, token by token, an enormous number of those options happen to contain a dash. It isn't choosing the em dash because it understands drama or emphasis. It's choosing it because, statistically, that's where the best writing in its training data tends to go.
One AI researcher put it in terms I found genuinely funny: real writers are annoyed that AI stole their signature move, and they're not wrong, because the models learned it from them in the first place. The em dash isn't an AI invention. It's a Virginia Woolf and Emily Dickinson and every well-edited magazine feature invention, absorbed wholesale and then deployed at a volume no individual human writer ever would.
And apparently this is why it's so hard to get rid of. Companies have tried tuning their models to obey "please, no em dashes" instructions, and it mostly works, sort of, some of the time. But researchers comparing AI text to human writing have found some models still use the mark more than three times as often as a typical human essay, even under direct instruction not to. One theory feels almost a little sad: telling a model not to use its highest scoring phrasing doesn't remove that phrasing from its sense of what "good" sounds like. It just makes the writing come out flatter as the model reaches for its second choice instead. The em dash has plenty of company on the list of tells, too. "Delve." "Myriad." "Nuanced." "Robust." Nobody has ever said "I'm going to delve into my inbox" out loud in the history of human speech, and yet there it is, over and over, because it scored well in the training data.
When I went digging into why nobody has properly fixed this, the answer read a bit like a case study in childhood formation. The habit doesn't get installed during the polite, well-behaved stage of training, the one where a model gets coached on manners, safety, and staying on topic. It gets installed much earlier, in the vast, unsupervised reading phase where a model first forms its raw sense of what good writing sounds like, long before anyone starts shaping how it should actually behave. Researchers comparing raw, uncoached models to their later, mannered versions have found the tendency already sitting there in the raw one, fully formed, before a single instruction was ever applied. Everything that happens after that, the fine-tuning, the "no em dashes" system instructions, a lab's own attempts at etiquette, is behavioural coaching layered on top of a personality that was mostly set at an earlier, formative stage. It can suppress the tic. It rarely erases it. And tellingly, one lab's coaching pushed the habit almost to zero while another's actually made it worse, so this isn't some law of nature the industry is helplessly stuck with. It's a choice about how much effort a company is willing to put into raising a particular habit out of its model, and plenty of them evidently haven't decided it's worth the effort.
There's a second, less flattering wrinkle sitting underneath all this. A meaningful share of the internet these models now learn from was itself written by an earlier generation of AI, em dashes and all. So the habit isn't simply failing to fade with time, it's being quietly fed back into the next round of training and reinforced a little further with each generation, the way a family trait can get passed down a touch stronger instead of bred out. Nobody designed it that way. It's just what happens when a machine's homework starts including its own previous homework.
The whole thing has gotten strange enough to make actual news. Back in May, Nike posted a tribute to tennis player Jannik Sinner that used a single em dash, and social media immediately accused the brand of letting AI write its copy. One dash. That's the level of scrutiny we're operating at now.
"We're not discovering a mark we'd been unfairly neglecting. We're watching one very specific, very literary habit get photocopied and pasted across the entire internet until it stopped looking literary and started looking automated."
A quick word on the pause hierarchy
While I was down this rabbit hole I got curious about where all our pause marks actually come from, because it turns out punctuation wasn't originally a writing tool at all. In ancient Greece and Rome, orators used early punctuation marks as cues for how long to pause when reading a speech aloud. The whole system started as stage directions for the voice, long before it became a rulebook for the page.
That legacy still shows up in how the marks behave today, roughly lined up by length of pause: the comma is the shortest breath, the semicolon about doubles it, the colon doubles it again, and the full stop closes the sentence entirely. The ellipsis does something different again. It isn't really a pause so much as an omission, a trailing off, a sentence that gives up on finishing itself. The colon has a specific job too: it promises the reader that what follows explains or expands on what came before. And the em dash, true to its "chameleon" reputation, can gatecrash almost any of those roles. It can do a comma's job, a colon's job, a semicolon's job, or a parenthesis's job, just with more visual space and more drama, which loosens the connection between the two halves of a sentence rather than tightening it. It's the least precise and most flexible mark we have, which probably explains both why writers love it and why a machine trained to sound polished would reach for it constantly.
Does this happen in other languages
I assumed, going in, that this would be a purely English-language crisis, and it mostly is, but the mark itself isn't uniquely ours.
Spanish uses a dash for a similar parenthetical purpose, though the spacing convention differs, with a space before the opening dash and none between the dash and the enclosed phrase. French and Italian do something similar, treating the dash almost like a bracket, with the closing dash often optional. Russian, interestingly, uses the dash more heavily than English does, partly because it stands in for the unspoken "is" or "am" that Russian sentences often omit entirely, a job English doesn't need a punctuation mark for at all. Chinese has its own version, the 破折号, which isn't actually the same character as our em dash. It's a full-width mark that takes up the space of two Chinese characters and signals an abrupt shift or an explanatory aside, doing a very similar conceptual job under completely different typographic rules. Japanese has a related but distinct mark used mostly to stretch out a sound rather than interrupt a thought.
So the impulse to mark an interruption or an aside seems close to universal. What varies wildly is the shape it takes, the spacing rules around it, and, presumably, whether the local AI chatbot is currently being begged to knock it off.
Are we just bad linguists
Which brings me back to the question I actually started with. Are we all just bad linguists who spent decades ignoring a perfectly good piece of punctuation, only for a robot to point out what we were missing? I don't think so, honestly. I think most of us simply never needed it. My design degree taught me to obsess over the space between letters and barely mention the dash, because in ordinary writing and ordinary speech, a comma or a full stop does the job just fine. The em dash lived mostly in novels and essays and the kind of prose most of us read but rarely wrote ourselves.
What's changed isn't that we've suddenly become more sophisticated writers. It's that a tool trained on our most polished, most edited, most "ten out of ten" writing has started handing that style back to us in every email, every LinkedIn post, every school assignment, at a volume and frequency no editor would ever allow. We're not discovering a mark we'd been unfairly neglecting. We're watching one very specific, very literary habit get photocopied and pasted across the entire internet until it stopped looking literary and started looking automated.
I still think the em dash is a lovely piece of typography: quietly, the shape of a capital M laid on its side, doing more work in a sentence than almost any other mark we have. I'm just not going to use one in this piece, on principle, and largely out of spite. Which, if you think about it, might be the strangest twist in this whole story. A punctuation mark nobody used to notice has become the one thing an entire generation of writers is now actively, self-consciously avoiding, purely to prove a very human point.
We were here first.
Emily Grinton is the co-founder of Emanda Ventures, where she spends more time than is healthy thinking about typography, technology, and the space between letters.
