Why writers are self-censoring em dashes now

Some writers have started deleting em dashes before they publish.

Not because the dash is wrong. Not because it suddenly makes a sentence unclear. Because readers have begun to treat it as a clue: was this written by AI?

It is an odd fate for a punctuation mark that has done useful work for centuries. An em dash can create a pause, set off an interruption, sharpen a turn in thought, or hold together a sentence that needs more movement than a comma allows. It is a real part of English prose. Plenty of distinctive writers use it beautifully.

But AI-generated writing uses em dashes a lot. It also uses them in predictable places: before a grand conclusion, around a soft qualification, or to give a generic sentence the shape of insight. Readers noticed the pattern. Then a practical observation hardened into a suspicion.

Now some writers worry that a dash will make their work look synthetic, even when it is exactly how they have always written.

That reaction is understandable. It is also a warning about what happens when people start writing around detection instead of toward meaning.

A punctuation mark became a shortcut for a larger feeling

The complaint about em dashes is rarely only about em dashes.

Most readers cannot reliably identify whether a paragraph came from a person or a model. They do not need to be able to. They are responding to a feeling: the prose seems too smooth, too evenly paced, too eager to explain itself, too familiar in its claims. The em dash becomes one visible feature they can point to.

That is how style tells work. A habit becomes noticeable when it appears alongside other habits. Overused section headings. A sentence that starts by naming a problem and ends by declaring a broad lesson. Three-item lists that sound balanced but do not add up to an argument. Phrases like "it is not just X—it is Y" deployed where a simpler sentence would do.

None of those moves is inherently bad. The problem is repetition without judgment.

An em dash is easy to notice because it is visually distinct. It has become a stand-in for all the moments when a reader feels the writer is following a template rather than making a choice.

Self-censorship solves the visible problem, not the real one

Removing every em dash can make a draft look less like a certain kind of AI output. It cannot make the draft more original, more specific, or more useful.

In fact, compulsively avoiding a punctuation mark can make prose worse. Writers replace a clean dash with a clumsy parenthesis. They split a sentence that needed one continuous motion. They trade a natural voice for a set of artificial rules designed to reassure an imagined detector.

The same thing happens when writers avoid words that models happen to use often, strip out any sentence with a parallel structure, or force every paragraph into an uneven rhythm. The work becomes a performance of humanity rather than an expression of it.

Readers do not ultimately trust a piece because it contains no suspicious marks. They trust it because it contains evidence of attention.

They can feel when an example comes from a real situation. They notice when a claim has limits. They remember when a writer names an experience more accurately than the usual category language does. Those signals are much harder to fake, and much more valuable than a dash-free draft.

The better question is whether the choice earns its place

There is a useful editing question hiding beneath the anxiety: why is this em dash here?

Sometimes the answer is good. The sentence needs an interruption. The second clause changes the meaning of the first. The dash creates a pause with more energy than a comma and less separation than a full stop. Keep it.

Sometimes the answer is less flattering. The writer wanted the sentence to sound dramatic. The thought was vague, so the punctuation was asked to make it feel weighty. The dash is covering for a transition that never became an argument.

That version deserves revision, but not because a dash is forbidden. It deserves revision because the sentence is asking typography to do the work of thinking.

This is a better standard for every stylistic choice. Why this metaphor? Why this opening? Why this piece of jargon? Why this conclusion? A strong draft can answer those questions with something other than "it sounded like writing."

AI fingerprints are usually context, not evidence

The temptation to treat one feature as proof of AI authorship will keep producing false positives.

People used em dashes before large language models. People will use them after today's model habits change. A writer's real voice may include patterns that happen to overlap with generated prose, just as a writer can avoid every fashionable AI tell and still produce something empty.

The more reliable signal is context.

Does the piece make a claim that depends on actual knowledge? Does it show where the writer is uncertain? Does it choose examples that fit the audience rather than filling a slot? Does it have an opinion with consequences, or only a polished summary of opinions everyone already has?

Those questions do not offer a quick verdict. They require reading. That is precisely why they matter.

The online rush to spot AI text is partly a response to a real problem: the internet is filling with work that feels unowned. But good readers should resist turning every stylistic habit into an accusation. And good writers should resist letting that anxiety narrow their tools.

Voice survives when the writer owns the decisions

Using AI does not automatically make a piece generic. Avoiding AI does not automatically make one human.

The difference is whether someone owns the consequential decisions in the draft. Who decided what the reader needed to understand? Who chose the example? Who noticed the claim was too broad? Who protected the strange, specific phrase instead of smoothing it into a familiar one?

AI can help a writer find repetitive phrasing, test an argument, or suggest alternative structures. It becomes a problem when the tool supplies the point of view and the writer merely approves its first plausible version.

That distinction matters more than the punctuation. A deliberate em dash in a draft with a real point of view is not an AI fingerprint. It is a tool used by a writer. A dashless post assembled from generic prompts is still generic.

Clarus is built around this kind of ownership. It is not here to make every draft sound uniformly polished. It is here to read closely: to flag the places where a thought becomes vague, where a voice gets smoothed away, and where a writer can make a more deliberate choice.

Use the em dash when it earns its place. Leave it out when another mark serves the sentence better. Just do not let a cultural panic about punctuation make the writing less like you.