The EM-dash Panic: Why We’re Ashamed of Our Own Shortcuts

Feature Article | Literature & Fiction

THE PREMISE

“ Writers have begun to fear AI-assisted prose, polished language and even a single punctuation mark. This editorial asks why we have become ashamed of our own shortcuts, what is fair and unfair in that suspicion, and why good writing should be judged by its ideas, its honesty, its clarity and its value to readers. ”

There is a particular kind of person who has emerged in the last few years.

This person opens an AI tool, describes what they want to say, receives a clean, well-organized draft in eleven seconds, and then spends the next three hours making it worse on purpose.

They swap “delve” for “look into.” They delete every em dash like a bomb-disposal technician. They insert a typo, then wonder if the typo looks too deliberate. They read the sentence “It’s not just X, it’s Y” aloud, gasp, and rewrite it as “It’s Y, and also somewhat X.” Finally they publish it, feeling not pride but the quiet dread of a man who has smuggled a watermelon through customs.

The Three-Hour Ritual

☑ Generate a perfectly good draft

☑ Replace “delve” with “look into”

☑ Delete every em dash

☑ Insert a typo, then doubt the typo

☑ Rewrite any sentence shaped like “not just X, but Y”

☐ Feel confident about the finished piece (still pending)

The writing was fine. The anxiety is the problem.

We have been here before

Every writing technology in history was first treated as a form of cheating. In Plato’s Phaedrus, Socrates worries that writing itself will weaken memory. Scribes grumbled about the printing press. Teachers insisted calculators would rot our minds, and spell-check was said to be making us illiterate. Editors, ghostwriters, speechwriters, and the friend who “just looks over” your cover letter have all been quietly shaping human prose for centuries.

c. 370 BCE

Writing. Feared as a threat to memory.

c. 1450

The printing press. Scribes grumble.

1970s

Calculators. Surely the end of arithmetic.

1980s–90s

Spell-check. A generation declared illiterate.

2020s

AI writing tools. Everyone becomes a forensic linguist.

Nobody opens a political memoir and demands a notarized affidavit that the leader personally typed every sentence. Nobody asks a CEO to confirm that the “heartfelt letter to our valued customers” wasn’t written by a committee, a comms team, and a lawyer who removed all the feeling. We have always known that most of what we read is the product of collaboration, polish, and delegation. The only new thing is that the collaborator is now a machine, and suddenly everyone has become a forensic linguist.

The performance of authenticity

Here is the strange part. The people who are most anxious about sounding like AI are often using AI the way a sensible person uses a good assistant: for structure, for a first draft, for the sentence they couldn’t land. Their ideas are real. Their experience is real. The point they’re making is theirs.

Yet they behave as though they’ve been caught shoplifting their own thoughts.

What’s happening, I think, is that we’ve confused two different things: the origin of the words and the value of the message.

The Question of Origin

Who typed it?
How long did it take?
Did anyone suffer?

The Question of Value

Is it true?
Is it clear?
Does it help the reader?

A recipe that works doesn’t become less delicious because a food processor chopped the onions. A bridge doesn’t collapse because the engineer used software. We judge most things by whether they do their job. Writing is the odd exception, where we’ve decided that suffering is the proof of quality. If it was easy, it can’t be good. If it was fast, it must be fake.

That belief is older than AI. It’s the same instinct that says a painting is worthless if it took only an afternoon, or that a student who finished the exam early must have cheated. It’s the Protestant work ethic wearing a cardigan.

Field Note · The Punctuation on Trial

The em dash did nothing wrong. Writers from Emily Dickinson onward have leaned on the dash long before any language model existed. Then chatbots began to favour it, the internet declared it a fingerprint, and in November 2025 one major AI developer even announced that users could instruct its chatbot to avoid em dashes. A mark of dramatic pause, once loved by poets, became a suspect in a lineup. Even its defenders admit the irony: this sentence has one — and nobody has called the police.

So why the jealousy?

Is it human nature, jealousy, or something else? It’s probably a bit of each.

1

Status

Fluency was once scarce, and scarce skills earn prestige. Cheap fluency feels like a shifting floor.

2

Tribalism

Humans love an in-group test, and “does this sound like a machine?” is the newest game in town.

3

Insecurity

The fear of being judged lazy, fraudulent, and the urge to sand every sentence smooth.

There’s a real status element. For a long time, fluent writing was a scarce skill, and scarce skills confer prestige. When a tool makes fluency cheap, people who built their identity around it feel the floor shifting. Some of the loudest “I can always tell when it’s AI” voices are really saying “I spent twenty years learning this, and I’d like that to still matter.” That’s an understandable feeling. It’s just not an argument.

There’s also plain tribalism. Spotting AI prose feels like detecting a counterfeit bill, and detecting things makes us feel clever. Never mind that the detection is wildly unreliable. People now accuse excellent human writers of being bots because they used a semicolon correctly. We’ve reached the stage where good grammar is circumstantial evidence.

50%+

misflagged

What the research found. In a 2023 Stanford-led study, popular AI detectors wrongly classed more than half of essays by non-native English speakers as machine-written, while accuracy on native samples stayed near perfect. The authors urged caution about using such tools to judge people’s work. Detection, in other words, can measure style while claiming to measure honesty.

And finally, there’s insecurity, which is the most human part of all. The anxious AI-assisted writer isn’t really afraid of readers. They’re afraid of being judged as lazy, fraudulent, or replaceable. So they overcorrect, sanding every sentence until it sounds like nobody at all.

What actually matters

Let me be the grown-up in the room for a moment, because satire only works if it’s pointing at something true.

The purpose of writing is communication. A reader wants to understand something, feel something, or decide something. If your article, email, book, or newsletter does that, then it did its job, regardless of whether a human, a machine, or a human-machine duet produced the sentences. The reader’s time was respected. The idea arrived intact.

But I’ll concede the fair point on the other side, because a thoughtful article should. Readers aren’t wrong to care about a few things. If someone is selling you “personal experience” they never had, that’s deception, with or without AI. If a piece is padded, generic, and says nothing in 2,000 words, that’s a quality problem, and the machine isn’t the only possible culprit. And in settings like academia, journalism, or contracts, where authorship carries rules, disclosure matters. Many journals and publishers now ask authors to disclose AI assistance and do not accept an AI system as an author, and that is a reasonable line to draw. Those are objections about honesty and quality, and they’re legitimate.

✓ Fair Objections

Inventing personal experience that never happened

Padding, vagueness and 2,000 words of nothing

Hiding AI use where rules require disclosure

✗ Weak Objections

“It sounds like AI, so it must be worthless”

“It used a semicolon, so it must be a bot”

“It was easy to write, so it can’t be good”

What isn’t legitimate is the blanket sneer: “This sounds like AI, therefore it’s worthless.” That’s not criticism. That’s a vibe check pretending to be an argument. Judge the work. If yes, the toolchain is none of your business. If no, it would have been bad with a quill pen too.

The Four-Question Test

I.

Is it true?

II.

Is it clear?

III.

Is it useful?

IV.

Is it moving?

Accept it and move on

So here is my modest proposal for the anxious writer.

Use the tool. Keep your ideas, your stories, and your judgment at the centre, because those are the parts a machine can’t supply. Edit until the piece sounds like you, not to dodge detection software, but because your voice is worth having. Be honest where honesty is expected. And then publish without the three-hour ritual of deleting em dashes.

A Modest Proposal for the Anxious Writer

Use the tool. Treat it like a capable assistant, not a confession.

Keep the centre human. Your ideas, stories and judgment are the parts a machine cannot supply.

Edit for your voice. Not to dodge detection software, but because your voice is worth having.

Be honest where honesty is expected. Follow the disclosure rules of your field.

Publish. Skip the three-hour ritual.

And to the reader: relax your grip on the magnifying glass. Ask whether the writing was worth your time, not whether it was typed by bare human fingers. You’ve been reading ghostwritten, edited, committee-approved, spell-checked, auto-completed text your whole life. You were fine.

The future of writing won’t be decided by who suffered most over the sentence. It will be decided by who had something worth saying, and said it well enough that you kept reading.

Which, if you’ve made it this far, I suppose I did. Whoever, or whatever, wrote it.

Feature Article | Literature & Fiction
Writing, authorship and the age of AI

Editorial Note

This editorial is a satirical and reflective commentary on the growing suspicion around AI-assisted writing, polished prose and individual punctuation habits. It does not argue against disclosure where institutions, publishers or professions require it, and it does not suggest that quality, honesty or originality are unimportant. It is directed at cultural attitudes, not at any individual, company or technology provider. Details about AI writing tools and detection research reflect published sources available as of October 2026 and may change.

References and further study

Plato. Phaedrus. Translated by Alexander Nehamas and Paul Woodruff. Indianapolis: Hackett Publishing, 1995.

Ong, Walter J. Orality and Literacy: The Technologizing of the Word. London: Methuen, 1982.

Liang, Weixin, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. “GPT Detectors Are Biased Against Non-Native English Writers.” Patterns 4, no. 7 (2023): 100779. https://doi.org/10.1016/j.patter.2023.100779.

Lindsay, Kate. “No—Em-Dashes Are Not a Sign of AI.” Embedded, March 10, 2025. https://embedded.substack.com/p/em-dashes-ai.

“OpenAI Says It’s Fixed ChatGPT’s Em Dash Problem.” TechCrunch, November 14, 2025. https://www.techcrunch.com/2025/11/14/openai-says-its-fixed-chatgpts-em-dash-problem/.