In Defense of the Em Dash

You've seen the replies.

Someone shares a post, and within minutes: "Spotted the em dash. AI wrote this." As if punctuation were a fingerprint. As if catching one makes you an editor.

It started a couple of years ago, when AI models got good enough to write whole paragraphs. The paragraphs had em dashes in them, and a certain kind of person decided this meant they'd cracked the code. Not because anyone changed the rules of punctuation. Not because the Chicago Manual of Style issued a retraction. The punctuation just showed up in machine-generated text, and the discourse needed a tell.

Large language models learned to write from a massive corpus of human writing. That corpus included em dashes — because writers have been using them for centuries. Emily Dickinson used them obsessively. So did Faulkner. So has every opinion columnist and feature writer who needed to interrupt a sentence with a sharper thought.

The em dash exists because a comma signals continuation, and sometimes a sentence needs to pause with a payoff. It's the punctuation equivalent of grabbing someone's arm mid-conversation and saying, "Wait — this part matters." It sets a different expectation. The reader leans in for a punchline.

That's not a flaw in the writing. That's a tool doing its job.

There's a Chuck Klosterman bit about people who tell you they like every kind of music "except country." His verdict: boorish and pretentious at the same time. The giveaway isn't the country part. It's the performance of taste in a setting that didn't ask for one.

The em dash backlash works the same way. It's never been a conversation about writing quality. It's a way for people whose only output is critique to position themselves as arbiters of what's real.

Calling out AI tells has become that performance. It says, "I'm perceptive enough to see behind the curtain." But perception without contribution is just heckling. If your entire engagement with someone's ideas is to flag the punctuation, you haven't actually engaged with the ideas.

This is punctuation I've used for two decades — long before any language model could string a sentence together. The idea that it's suddenly suspect because a model also reaches for it is absurd.

Some AI tells are problems. This isn't one.

There are patterns in AI-generated text that make writing worse. Triple stacking — listing three parallel items in every paragraph — gets monotonous fast. Negative parallelisms ("not this, but that") become a crutch when every other sentence runs the same template. Summary sentences that restate what the paragraph already said treat the reader like they can't hold an idea for three lines.

(I'm aware I just stacked three of them. The trick is when you don't notice.)

These patterns interrupt the reading experience. They make prose feel mechanical.

The em dash doesn't do that. An em dash mid-sentence doesn't break your concentration or make your eyes glaze over. At worst, overuse makes prose feel breathless. But overuse of anything is bad — semicolons, exclamation points, the word "however." That's an argument against excess, not against the tool itself.

The strongest version of the case against em dashes goes further: language models do produce them at higher per-word rates than the average writer, so the punctuation IS a probabilistic signal.

Granting that — the false positive rate is enormous. Plenty of writers have always used em dashes liberally, and dragging them along as collateral damage costs the language a tool it hasn't done anything to deserve. The signal is fading anyway. AI labs have been tuning models toward fewer em dashes, which means the detectives are hunting yesterday's pattern while the actual problems (generic praise, vague attributions, inflated significance) keep doing their thing in the background.

This drives me nuts.

Evaluate the content, not the toolchain

Nobody writes in a vacuum anymore. AI is part of how people draft, edit, think, and communicate. That's true for LinkedIn posts, emails, pitch decks, and probably half the Slack messages you read today.

The line between "human-written" and "AI-assisted" is already blurred past the point where policing it makes sense.

So what's left? Evaluate the content on its merits. Does it say something worth reading? Does it connect with the audience it's meant for? Does it move an idea forward or start a conversation that matters?

Slop is slop. Content that says nothing, adds nothing, and exists only to fill a feed is bad — and it's bad whether a human typed every word or a model generated it in three seconds. But the presence of an em dash is not what makes something slop. The absence of an idea is.

The Scarlet M

At Postworthy, AI is part of the pipeline. We don't hide it, and we don't lead with it, because the goal was never "use AI." The goal is to help people communicate ideas that matter and make their expertise visible. AI is the tool. The ideas are the point.

The em dash has become a kind of scarlet letter, where spotting one is supposed to delegitimize everything around it. That's a lazy way to engage with content. It requires zero understanding of what was written and zero effort to contribute something better.

If you communicate about your work — pitching, presenting, writing, telling the story of what you do so you can keep doing it or do more of it — you shouldn't spend a single minute worrying about whether your punctuation will get you accused of using AI.

Use the em dash. Use the tools that help you communicate clearly. Spend your energy on the ideas, not on performing purity for people who were never going to engage with your work anyway.

The em dash has been doing its job since long before any language model existed. It'll keep doing its job long after the discourse moves on.

Keep using it.

Sources

• Chuck Klosterman, "Contrary to what you may have heard from Henry Rollins…" — Sex, Drugs, and Cocoa Puffs: A Low Culture Manifesto (2003)