The Em Dash Is Not an AI Tell. It’s Punctuation.

The em dash is not an AI tell. It is a piece of standard English punctuation that predates large language models by roughly two centuries, and the panic over it is making web content worse, not safer.

I am writing this knowing it probably will not rank. The phrase “em dash ai tell” is owned by opinion essays and culture-war takes, and a Charlotte web agency is not the expected voice in that conversation. I am publishing it anyway because it is correct, and because we have personally made the mistake I am about to tell you to stop making.

The em dash (—) is correct punctuation, not evidence of AI authorship. AI detectors weigh statistical patterns like perplexity, not your punctuation. The real cost of stripping em dashes out of everything you publish is not safety. It is homogenization. You are sanding your brand voice down to match everyone else who panicked.

Is the em dash actually an AI tell?

No. The em dash is not a reliable signal of AI authorship, and treating it as one is a category error. AI models use the em dash heavily because it is correct in many situations, and the models were trained on edited, published English, where it appears frequently.

Read that again. The reason machines produce em dashes is that em dashes are right. The tool learned good punctuation from human writing, and now we are reading the punctuation back as a crime. That is backward.

The honest version of the fear is this: people saw a lot of ChatGPT output, noticed it used em dashes freely, and built a folk rule that an em dash equals a robot. Folk rules spread fast because they are easy. This one is wrong.

Practical rule: If a punctuation mark is correct, the fact that a machine also uses it correctly does not make it suspect. Edit for clarity, not for innocence.

We strip em dashes out for clients, and that is the mistake

Here is the part most agencies will not admit. At Four Eyes, we have been hypersensitive about removing em dashes from content. I have asked clients to have every em dash pulled out, every time, as a reflex. We have spent real billable effort doing find-and-replace passes to scrub a correct punctuation mark out of perfectly good copy.

It is getting out of control. We lost sight of what this punctuation is, what it means, why it is useful, and why we should not abandon it just because a model might know grammar and punctuation better than the person who is afraid of it.

So this is me correcting the record, including my own behavior. We use AI at Four Eyes for the parts it is genuinely good at. We use it to rough out wireframes during web design and for SEO and SERP analysis, so we write about topics that actually matter to a client’s audience. That is a legitimate use of the technology. Deleting correct punctuation to dodge a detector is not.

Practical rule: Use AI to decide what to write about and how to structure it. Do not use a detector phobia to decide how to punctuate it.

What an em dash actually is, per the style authorities

The em dash is one of three horizontal marks that people constantly confuse: the hyphen (-), the en dash (–), and the em dash (—). They are not interchangeable, and the difference is not decorative.

• Hyphen (-): joins words and parts of words. “Well-known agency.” “Long-term plan.”
• En dash (–): spans ranges and connects related things. “2019–2024.” “The Charlotte–Raleigh corridor.”
• Em dash (—): marks a break in thought, sets off a parenthetical with emphasis, or introduces a summary. “We tried three tools, none of them helped.”

The Chicago Manual of Style treats the em dash as standard punctuation with defined jobs. It can replace commas when you want a stronger pause, replace parentheses when you want the aside to feel emphatic rather than buried, and replace a colon when you want a faster setup before a summary. The Chicago Manual of Style describes the em dash as the most versatile punctuation mark for exactly these reasons.

None of those jobs is new. None of them came from a model. A writer in 1890 used the em dash the same way a clean LLM output does today, because the rules had not changed.

Hyphen vs. en dash vs. em dash: what each one is for

Practical rule: If you cannot tell whether you want a hyphen, en dash, or em dash, that is a punctuation gap to fix, not a reason to delete all three.

The “width of a capital M” myth, corrected

You have probably been told the em dash is “the width of a capital M.” That is the tidy explanation everyone repeats, and it is wrong.

The “em” is a typographic unit equal to the current point size of the type. In 12-point type, one em is 12 points wide. It is named after the letter M because, in many old metal typefaces, the M was cast on a square body roughly that wide. The unit is tied to the point size, not to the actual rendered width of the glyph M in whatever font you are using. In most modern fonts, the capital M is narrower than a full em.

So the em dash is “one em long,” and an em is a unit of measure. The M is the origin of the name, not a live measurement. Small correction, but it is the kind of detail that separates people who actually set type from people repeating a meme.

Dash usage peaked in the 1860s, not in 2023

If you think heavy dash usage is an AI invention, the historical record disagrees with you. Analysis of large book corpora shows that dash usage in English-language books climbed through the 1800s, reached a peak around the 1860s, then declined for over a century before any of this. Long-running analysis of digitized book text tracks that arc clearly.

Emily Dickinson built an entire poetic voice on the dash in that era. Victorian prose leaned on it constantly. The mark you are scrubbing out of a landing page was a defining feature of nineteenth-century writing.

The dash did not appear with ChatGPT. It came back into heavy rotation, sure. But “a punctuation mark is being used a lot again” is not a scandal. It is a return to a tool good writers always had.

How do AI detectors actually work? (Not by counting dashes.)

AI content detectors do not flag em dashes. They estimate how statistically predictable your text is. The two concepts that matter are perplexity and burstiness.

• Perplexity: how surprised a language model is by your next word. Low perplexity indicates very predictable text that reads as machine-like. High perplexity means more surprising word choices, which read as human.
• Burstiness: how much your sentence length and complexity vary. Humans write in bursts. Long sentence, short one. Machines tend to be more even.

Notice what is not on that list. Punctuation choice. A detector models the statistical shape of your writing, not audits your dashes. You can write a paragraph stuffed with em dashes that reads as obviously human, and a dash-free paragraph that pings every detector because it is flat and predictable.

This is also why detectors are unreliable in both directions. They produce false positives on careful human writing and false negatives on lightly edited AI writing. Chasing them by changing punctuation is like trying to pass a metal detector by changing your shirt color. Wrong variable.

Practical rule: If you want your writing to read as human, vary your sentence rhythm and make real, specific word choices. Punctuation cosmetics do nothing to the signal detectors measure.

The model makers are already dialing the em dash back

Here is the news peg that makes the whole find-and-replace industry look silly. The companies building these models have started tuning down the em dash habit because users complained about it. OpenAI has publicly acknowledged adjusting the behavior so the model honors a user’s request to avoid em dashes.

And if you genuinely do not want them in a given draft, you do not need a scrubbing pass. You add one line to your prompt: “Do not use em dashes.” That is the whole fix. OpenAI’s product updates note that the model now follows instructions to avoid em dashes more reliably.

So the supposed “tell” is suppressible at the source with a sentence, and the people who built the tool are already turning the dial. The frantic post-hoc deletion across thousands of pages is solving a problem that barely exists and degrading the writing in the process.

The real danger is homogenization, not detection

Here is the thesis: AI should make your SEO better, not blander. The threat from AI in content is not that a detector outs you. The threat is that everyone uses the same tools the same timid way, and the entire web converges on one beige voice.

When you delete a punctuation mark because you are scared of a machine, you remove a tool of emphasis and pacing. The em dash creates a specific rhythm. The interruption. The aside that hits harder than a comma would. Strip it out wholesale, and your prose gets flatter and more uniform.

Now multiply that by every marketing team running the same defensive playbook. The web fills with content that was edited to look like nobody wrote it. That is the actual failure. Not a flag in some detector dashboard. A brand voice that sounds exactly like its competitors.

A distinct voice is a ranking and conversion asset. Search engines and AI answer engines are increasingly rewarding content that demonstrates real expertise and a point of view. Our work on technical SEO and answer engine optimization keeps pointing at the same conclusion. The pages that get cited and quoted are specific and confident, not laundered to be inoffensive.

Practical rule: Defend the parts of your writing that make it sound like you. Punctuation that carries emphasis is part of that, not a liability.

When should you actually remove an em dash?

Sometimes. The argument here is against reflexive deletion, not against editing. There are real reasons to swap an em dash for something else, and none of them involve fooling a detector.

• Overuse in a single passage: three em dashes in two sentences is a structure problem. Rewrite for clarity. The fix is fewer dashes because the writing is cleaner, not zero dashes out of fear.
• House style demands it: some brands genuinely standardize on commas and parentheses. That is a defensible editorial choice. “An AI might use it too” is not the reason. Consistency is.
• Accessibility and rendering: Some platforms mangle the character, and screen readers handle the em dash inconsistently. If your CMS turns it into a mojibake question mark, fix the encoding, do not abandon the punctuation.
• Wrong dash for the job: people type an em dash where an en dash or hyphen belongs. Correct the mark, do not delete the concept.

Every one of those is an editorial judgment about the reader. That is the only standard that counts. Edit for the person reading, never for the imaginary auditor counting your dashes.

A sane workflow for publishing in the AI era

Here is what we do now, after walking back our own dash phobia. This is the process for any team using AI as part of content production without letting it flatten the work.

1. Use AI for research and structure, not final voice: SERP analysis, topic clustering, and wireframes are where it earns its keep. We use it to find what to write about, then humans write the angle.
2. Edit for perplexity and burstiness, not punctuation: if a draft reads flat, the fix is specific examples, varied sentence length, and a real opinion. Not a find-and-replace on the dash key.
3. Set the prompt up front: if a client’s house style bans em dashes, add “Do not use em dashes” to the prompt. Solve it once, at the source.
4. Check encoding, not just text: make sure your WordPress setup renders the em dash as a clean character across browsers and screen readers.
5. Keep a real voice: the goal is content that sounds like a specific human at a specific company. That is what the asset detectors cannot fake, and competitors cannot copy.

Short version. The em dash is correct, the detectors do not care about it, the model makers are already tuning it down, and the only thing wholesale deletion accomplishes is making your content sound like everyone else’s. We learned the hard way by doing the wrong thing for clients, and we are done with that.

Sources

• The Chicago Manual of Style treats the em dash as standard, versatile punctuation that can replace commas, parentheses, or a colon.
• OpenAI has adjusted ChatGPT so it more reliably follows a user instruction to avoid em dashes.