Deleting what never needed to be deleted.
I was writing one of my essays and I deleted an em-dash.
I didn’t delete it because it was wrong.
Nor because the sentence did not need it.
I deleted it because… well… it looked like AI.
The sentence was fine. An em-dash would help the reader. And I knew it.
But I deleted it anyway. And used a semi-colon.
This is happening with little fanfare, across many drafts, in many documents no one else will ever see. Writers are auditing their own work for patterns that resemble machine output. They find something. They remove it. Replace it. And move on.
The tools being set down are not obscure ones.
The em-dash. The triad of short sentences. The single-word paragraph for emphasis. The colon that opens into a list. The pivot that arrives without warning. These are not the tics of careless writers. They are the instruments of careful ones.
You used them, before the question of resemblance entered the room.
These tools are not new. What is new is the specter of AI.
These tools now carry a second meaning, and that meaning is: suspect.
Consider what actually happened.
Language models were trained on text. Enormous amounts of it. They absorbed the patterns of published writing, including the patterns of writers who knew what they were doing. The models learned emphasis and rhythm. They learned to reach for the em-dash, the short sentence, and the well-placed fragment. Then they overused everything.
For AI, repetition is easier than restraint.
They learned the gestures without the reasons. They learned that short sentences create emphasis without learning that emphasis requires tension. They reached for the tools constantly, indiscriminately, until the tools began to feel like a signature.
And whose problem is that?
It has become ours.
We are now revising our sentences to avoid resembling a system that was trained on sentences like ours. We are making our writing less like itself, so it does not appear to be the output of a process that made itself more like our writing.
Read that again slowly.
The model learned from us.
We are now learning from the model.
What it learned was our habits.
What we are learning is to hide them.
This is not caution. It is not professionalism. It has a simpler name.
We changed our writing because a machine wrote badly and the reader might not know the difference. We accepted, without much ceremony, that the machine’s failures were now our problem to manage.
That is surrender.
The em-dash did not change.
Readers changed.
And we changed to meet the reader’s new suspicion.
The machine, which has no awareness of any of this, continues exactly as before.
That is the absurdity.
The writer must now surrender the tools the machine learned to imitate.
Somewhere, a machine is writing. It is reaching, right now, for the em-dash.
The semi-colon I used keeps changing to an em-dash. And back.
I still have not decided what to do about that.
Top comments (0)