I’ve been internet acquaintances with Mignon Fogarty, also known as Grammar Girl, for many years now. (We share a literary agent, in fact, which is perhaps the most New Yorker–coded thing I’ve ever typed.)
As such, I was more than happy to talk to Mignon about Face with Tears of Joy for her always fascinating podcast. That episode has been out for a couple of months now (you can find it here, on YouTube, or wherever else you get your podcasts), but we also recorded a bonus episode about the hoary subject of books and book history. I am happy to report that that second episode has now been published to YouTube and all of the usual places, so head over there to watch and listen. (And also to marvel at my webcam, which insisted on focusing on the bookshelves behind me rather than on my face. It has a sense of irony, at least.)
There is a new AI controversy in town: chatbots are ruining the em dash.
Over the past year or so, a number of people have decided that texts created by generative AI applications contain more em dashes than might otherwise be expected. The complainants come from LinkedIn,1 Reddit,2 Instagram,3 and beyond. There’s even a thread on a forum operated by OpenAI itself,4 the company which owns ChatGPT, whose participants bemoans the bot’s excessive and apparently unstoppable use of em dashes.
What, though, is an em dash? The Shady Characters book has a chapter dedicated to dashes, but for now it’s enough to know that the em dash is one of a family of dashes used for different typographical purposes.
The hyphen (-) is the shortest dash, and is often used to join compound terms. Note the lack of spaces: “Moby-Dick.”*
The en dash (–), which is slightly longer, is used to join ranges of numbers (1–10) or dates (2010–2025), and, mostly in British English, to mark a break or parenthesis in a sentence – and in this mode, it normally has a space on each side. The en dash is also sometimes used to hyphenate compounds which themselves contain compound terms. For the keen-eyed, there’s an example elsewhere in this post.
The em dash (—) is the longest of the conventional dashes. It is named for the em, a typographic unit of size, which was once held to be the same width as an ‘M’ from its parent typeface, but which is now synonymous with the size of the typeface itself. A 16-point font will have a 16-point em, and its em dash will be 16 points long. (An en, incidentally, is half the width of an em.) The em dash is used by some novelists to introduce a change in speaker when recounting direct speech, although the practice varies from writer to writer and from country to country. Here’s a contrived example (contrived, that is, because Melville used quotation marks in Moby Dick):
— Masthead, there! Look sharp, all of ye! There are whales hereabouts! If ye see a white one, split your lungs for him!
— What do you think of that now, Flask? ain’t there a small drop of something queer about that, eh? A white whale—did ye mark that, man?5
More commonly than this, however, the em dash is used in American English for the same purposes as the en dash in British English—that is, to set off a parenthetical clause or to mark a break in a sentence. In this role, it’s normally used without spaces.† And this, finally, is the mark to which large language models are alleged to have laid claim.
Back to the flap at hand. The complaint is that many of the LLM-driven chat services that currently occupy the public consciousness‡ use em dashes so frequently that they have become a hallmark of AI-generated text. A viral clip from a podcast called LuxeGen (you can watch it here),3 in which the hosts discuss a rebrand of a clothing label called Pretty Little Thing, gave the phenomenon a name: the “ChatGPT hyphen”.
The odd thing from my perspective is that documentary evidence of the ChatGPT hyphen is quite thin on the ground. Opinion about the ChatGPT hyphen, on the other hand, is rampant. Quite apart from the blog posts6 and newspapers articles7 recapitulating the issue, a handful of technically-minded em dash–phobics have come up with ways to actively fight back against it, with the two most striking efforts both appearing in May this year.
First came the “am dash”. As you can read at theamdash.com, the idea is that human writers should not use em dashes at all, but instead the “am dash” — a sort of elongated tilde, or a dash with one end turned up and the other turned down — to distinguish themselves from AI chatbots. The brainchild of an Australian ad agency called Cocogun, the am dash is, so the website claims, “unmistakably human, unusable by AI”.8
Putting the am dash into practice is somewhat involved. There’s no standardised number or code that identifies this novel mark of punctuation, so the only way to use it is to install one of two custom fonts (“Areal”, based on Arial, and “Times New Human”, based on Times New Roman) and then, rather than typing an em dash, type in “am-” instead. Areal or Times New Human, as appropriate, will recognise this special invocation and replace it with an am dash. It’s clumsy, but if you have the luxury of choosing your own typeface, it works.
The thing is that any chatbot worth its salt can very easily make the same substitution. I asked Google’s Gemini: “Please write me a sentence illustrating the use of the em dash, but instead of the em dash itself, please use the characters ‘am-’.” This was the response I got:
I wanted to go to the store am-but then I remembered I forgot my wallet.
When I copy-and-pasted this into a word processor using Areal, this was the result:
llustrating the use of the am dash in Microsoft Word. (Image by the author.)
Admittedly, there’s a spurious space before am dash, but I think the point stands. The am dash is more social commentary than a valid antidote to the ChatGPT hyphen.
theamdash.com was followed by noemdash.com. (Ironic, I think, that both URLs could use a hyphen or two to make them easier to parse.) This, in turn, was the product of Lior Grossman, a self-described “serial parallel entrepreneur”, who had, he wrote, “lost it with ChatGPT”:
For probably the 100th time, I patiently asked it (okay, maybe less patiently each time) to STOP using em dashes. I even spelled it out clearly in the instructions: “Do NOT use em dashes (—). Just hyphens!”. What did ChatGPT do? It nonchalantly sprinkled even MORE em dashes, mocking me, as if my instructions were just a friendly suggestion 🤯9
Grossman’s response was to build noemdash.com, a tool which scrubbed texts of em dashes and replace them with spaced hyphens. “Get text that doesn’t suck”, it boasted. And to be fair to it, noemdash.com does what exactly what it promises to do: it removes not only em dashes, but also semicolons, “curly” or “smart” quotes, and also the hidden watermarks added to text by some AI systems.
Unfortunately, both Grossman and Cocogun have missed the point: we should not be picking on the em dash. To anyone with a passing interest in writing or typography, the em dash is not a malign mark: on the contrary, what an em dash most reliably signals is that a writer or a proofreader or a typographer cared about their work.
With the caveat that there are no rules in writing and typography, only conventions, there are many situations where an em or en dash is inarguably the best tool for the job. There are other marks, too, which are, for some definition of correctness, more correct than their lo-fi equivalents. In an ideal world, possessives would come with apostrophes (’) rather than primes (′). Direct speech would live between inverted commas (“ ”) and not double primes (″ ″). Independent clauses would be gracefully united by semicolons rather than hastily epoxied with commas.
If you are writing for publication, then, or in a formal register, or you just want to do the right thing by your words, then you should absolutely be reaching for the em dash and its comrades. There is no shame in this. And if you are evaluating a piece of writing in those same contexts, then by extension the em dash is no more helpful in identifying a chatbot’s ersatz spiel than any of the other, mostly-discredited tricks that have been claimed to be able to do so.10 In this light, the am dash and noemdash.com look almost parodic: theamdash.com asks that you confine yourself to one of only two custom fonts in order to avoid looking like an chatbot, while noemdash.com goes to extraordinary lengths to excise any sign of conscientious punctuation or typesetting from a piece of writing.
Of course, this isn’t to denigrate writing which is, by that same definition of correctness, incorrect. Real life comes for us all! It takes a few extra taps or clicks to find a curved quotation mark, and who has the time for that? The Chicago Manual is right there on the shelf but the kettle is boiling or the kids are crying. Bluesky fails to convert a well-intentioned double-hyphen to an em dash. There are a million reasons why we dash off texts that aren’t quite as refined as they could be.
If I have a conclusion, it is something like this: raging that a chatbot’s words are typeset better than a human’s is futile. LLMs are trained on a huge variety of texts, many of them from a bygone age when publishing anything at all meant having it first edited, then copyedited, then proofread, and finally typeset. If an chatbot is going to learn anything at all from the centuries’ worth of pirated books it has been fed, it will be where to use an em dash. Better, perhaps, would be to look critically at our use of AI in the first place. If you don’t like being outed as a ChatGPT user, maybe the solution isn’t to wreck its typography — it’s to stop using it in the first place.
Weber-Wulff, Debora, Alla Anohina-Naumeca, Sonja Bjelobaba, Tomáš Foltýnek, Jean Guerrero-Dib, Olumide Popoola, Petr Šigut, and Lorna Waddington. “Testing of Detection Tools for AI-Generated Text”. International Journal for Educational Integrity 19, no. 1 (December 2023): 1-39. https://doi.org/10.1007/s40979-023-00146-z.
As you will have seen in this post and in others here at shadycharacters.co.uk, I quite enjoy using spaced em dashes. This is contrary to essentially every style guide. Don’t do it. ↢
If you won, congratulations! Look out for an email from me with your audiobooks.com code for Face with Tears of Joy. If not, don’t despair — ‘😂’ is available at all good bookshops.
*
In more detail: I arranged the names of all entrants in a text file, then used random.org to pick two numbers between 1 and the total number of lines in that file. ↢
To receive your prize, you’ll need an account at audiobooks.com or audiobooks.co.uk. (Other localised versions of audiobooks.com may work, but unfortunately I can’t guarantee it.)
The contest will close at noon UK time on Sunday 10th August 2025, so make sure you enter before then. After that I’ll pick four winners from the list of all unique entrants, and I’ll get in touch with a code to access your copy via audiobooks.com. See below for terms and conditions, and good luck!
Update: The competition is now closed! I’ll be in touch with the winners in due course.
