The Em Dash Is Not the Problem

After I published the first article on Engineering the Conversation, my brother offered feedback from a reader’s perspective. One point stuck with him enough to mention directly: I should probably stop using the em dash. It “felt AI.”

A few months earlier, I wrapped up a writing series for my company channel and created a companion guide to go with it. The Communications team reviewed it and sent back suggested edits. One stood out.

Remove the em dash.

Not because it was grammatically incorrect. Not because it disrupted readability. Remove it because the em dash has become a telltale sign of AI-generated writing.

Two separate moments. Months apart. Different audiences. Same conclusion.

And in both cases, the feedback came from a good place. Nobody was accusing me of cheating or trying to flatten my voice. They were trying to help me avoid a perception problem. The em dash, somehow, had become suspicious.

The irony is I’ve been using the em dash since high school.

Not casually. Not accidentally. I mean leaning on it, the way some writers lean on commas or line breaks or silence. It gave me rhythm. It let me pivot mid-thought. It made my writing feel like thinking instead of reporting.

And now, apparently, it makes me sound like AI.

If you spend any time in writing circles lately, you’ll hear it: the em dash is a giveaway. A quiet signal that what you’re reading might have been machine-assisted, machine-polished, or machine-generated outright. The advice follows quickly: avoid it, reduce it, replace it.

Kick your old friend to the curb.

Photo by Joachim Schnürle on Unsplash

That’s the part I’m not buying.

Because the em dash isn’t new. It isn’t artificial. It isn’t some stylistic mutation born in the age of large language models. It has a lineage. One that stretches back to medieval scribes who used long horizontal strokes to mark pauses in handwritten text. It evolved through the constraints of typewriters, where writers were forced into the double hyphen workaround. And somewhere along the way, it became something more expressive than functional.

Ask Emily Dickinson.

Her use of the dash wasn’t mechanical. It was musical. Emotional. Ambiguous in a way that made meaning feel alive rather than fixed. The so-called “Dickinson Dash” didn’t just connect clauses; it created tension. It invited interpretation. It let the reader hover in uncertainty for just a moment longer than grammar typically allows.

That’s not a bug. That’s the point.

So what changed?

AI didn’t invent the em dash. It optimized it.

Language models are trained on vast amounts of clean, modern prose: essays, articles, think pieces, marketing copy. In that ecosystem, the em dash shows up as a reliable tool. The punctuation equivalent of a session musician who can play anything competently and nothing memorably.

So the model learns: when in doubt, use the em dash.

And use it often.

What we’re seeing now isn’t corruption. It’s compression. A stylistic move that once had range and personality is being reduced to a handful of predictable functions: soft pivot, added emphasis, quick aside. The em dash becomes a shortcut to sounding thoughtful.

That’s why it’s starting to feel generic.

Not because it’s artificial. Because it’s overfit.

And here’s where the conversation usually goes off the rails. Instead of asking how to use the em dash better, we start asking whether we should use it at all. As if the presence of a punctuation mark is the problem, rather than the way it’s being deployed.

The real issue is sameness.

AI doesn’t just generate text. It converges toward what is statistically “good.” Clear. Coherent. Polished. And in doing so, it narrows the stylistic spectrum. The edges get sanded down. The weirdness gets averaged out.

The em dash just happens to be one of the tools caught in that gravitational pull.

So now we’re left with a strange dilemma: if you write the way you’ve always written, you risk being mistaken for a machine. And if you change your style to avoid that, you’re letting the machine redefine what “human” writing looks like.

That feels like the wrong trade.

Because the alternative isn’t to abandon the em dash. It’s to reclaim it.

You may have noticed I’ve barely used one in this piece. That was deliberate. I wanted to feel what it was like to earn each one back.

Use it in ways AI doesn’t. Let it interrupt, not just clarify. Let it create ambiguity instead of resolving it. Let it carry emotional weight, not just structural convenience. And maybe most importantly, use it sparingly enough that it actually registers when it appears.

In other words—make it a choice again.

This is the broader shift I think we’re underestimating. AI isn’t ruining the tools of writing. It’s standardizing their usage. It’s turning style into pattern, and pattern into expectation.

And once something becomes expected, it stops being expressive.

Which brings me to the hill I’m actually willing to die on: the ellipsis.

Because if the em dash has been optimized into predictability, the ellipsis still resists.

It doesn’t clarify. It withholds.

It doesn’t resolve. It lingers.

It doesn’t tell the reader what to think. It invites them to finish the thought themselves.

That’s a very different posture. And it’s one that AI, by default, tends to avoid. Language models are designed to complete ideas, not leave them open. To provide answers, not create space. Even when they use an ellipsis, it often feels decorative. A stylistic nod rather than a genuine pause.

But a well-placed ellipsis does something else entirely. It introduces doubt. Or tension. Or subtext. It signals that what’s unsaid matters just as much as what’s on the page.

And that’s harder to replicate.

Not impossible. But harder.

So maybe that’s the line we should be paying attention to. Not which punctuation marks are “safe” or “suspicious,” but which ones still allow for intentional incompleteness. Which ones create room for the reader instead of closing the loop for them.

Because that might be the real distinction emerging in this moment.

Not human vs. machine.

But resolved vs. unresolved.

Polished vs. alive.

And if that’s true, then the future of writing isn’t about avoiding the fingerprints of AI. It’s about leaning into the things that resist optimization. The choices that don’t just convey meaning, but complicate it.

The em dash isn’t the problem.

But how—and why—we use it?

That’s still very much ours to decide.