Modern AI writing style has a sound. You can hear it in the rhythm, the polish, and strangely, in the punctuation. Now, the em dash has become its brand, signaling its demise.

The em dash didn’t die of natural causes. It was murdered by overfamiliarity.

For about a century—give or take the typewriter era that struggled to print it—the em dash (—) functioned as a stylistic fingerprint. Writers reached for it when a comma felt timid and a period felt authoritarian. It let sentences think out loud.

Then large language models arrived.

When punctuation became a signal of AI writing style

Machines didn’t invent the em dash habit. Humans did. But models trained on mountains of human writing noticed something before we did: the mark correlates strongly with voice. Not grammar. Not clarity.

Voice.

Specifically, the modern essayistic voice—confident, conversational, reflective. The sound of a writer thinking in real time. A silent “um.” The pause before the thought lands.

So it’s no surprise the models learned to reproduce it. Faithfully. Reliably. Statistically.

Almost overnight, the em dash stopped signaling human cadence and started signaling machine cadence.

Why the em dash sounds human (and now doesn’t)

Here’s the paradox: AI didn’t change punctuation; it changed semiotics. The mark itself is unchanged, but its interpretation has flipped. Frequent em-dash usage once implied a person trying to sound natural. A human being human for other humans.

We’ve wandered into a linguistic uncanny valley.

Now it implies a system trained to imitate one, and a clear sign that someone has ventured into the realm of AI-generated content.

How readers detect AI-generated text without software

You can watch the immune response in real time. Writers avoid the scarlet letter of the em dash. They retreat to periods, commas, and abrupt fragments. Some adopt strict subject-verb-object sentences, like filing a police report about their own thoughts.

Once style becomes mass-produced, it ceases to signify effort.

We’ve seen this before—the same cycle that ended

the ukulele cover song
the sepia filter on Instagram
the lowercase minimalist brand voice
the blog post opened with a rhetorical question

They didn’t exactly become bad, more like they slid into cheapness.

Language is part communication and part authentication. When the authentication layer breaks, writers migrate.

The Victorians ornamented prose until the telegraph punished verbosity. Hemingway reacted to Henry James. Twitter reacted to blogging. Now humans react to machines that learned the blogging voice better than bloggers did.

The em dash became collateral damage.

It still performs the same function, but readers no longer perceive cognition in motion. They perceive probability distribution—a red flag that the text may be synthetic.

The future of human writing after AI

Which suggests a future in which human writing won’t outrun AI by becoming more elaborate. It will become more literal, more idiosyncratic, less polished—not because writers got worse, but because polish was automated.

Authenticity will live in friction:

typos that survive revision
rhythms that refuse averaging
metaphors too specific to generalize

Writing that refuses statistical convenience.