I wrote a post a while back about the em dash. About how LLMs have embraced it so thoroughly that now, whenever a human uses one, it looks like AI-generated text.
I stand by that observation. But I’ve been thinking about what comes after the observation.
I’ve started second-guessing my own writing.
Not because my writing changed. Because the world’s perception of writing changed.
I’ll catch myself mid-sentence, about to use an em dash—and then I pause. Will this look like AI wrote it? I’ll structure my thoughts clearly, with a natural flow from premise to conclusion, and then wonder if that’s too… organized. Too suspiciously coherent.
It’s absurd. I’m editing my own voice to not sound like machines that learned from us.
LLMs were trained on human text. Billions of words, written by actual people over decades. The reason AI sounds a certain way is because humans wrote that way first.
The em dash? Some humans loved it long before GPT existed. Clear structure? That’s just good writing. Thoughtful transitions between ideas? That’s called knowing how to communicate.
If an AI sounds like a well-written human, that’s because it absorbed patterns from well-written humans. The causality runs in one direction. They learned from us.
So why am I contorting myself to sound unlike something that learned to sound like me?
There’s a growing paranoia around writing now.
People consciously making their prose messier. Adding typos on purpose. Avoiding certain sentence structures. Hedging in awkward ways because clarity has become suspicious.
We’re self-sabotaging to prove our humanity.
And the irony is thick: the anxiety about sounding like AI is itself a kind of inauthenticity. We’re not writing naturally anymore. We’re writing defensively.
So here’s my permission slip. To myself, and to anyone else feeling this.
Write how you write.
Use the em dash if that’s how your brain works. Structure your thoughts if structure serves the idea. Be clear if clarity is what you’re after.
Don’t apologize for sounding coherent. Don’t introduce artificial roughness to prove you have a heartbeat.
If your authentic voice sometimes overlaps with how a language model sounds—so what? That overlap exists because the model learned patterns from authentic human expression in the first place.
You are not derivative of AI. AI is derivative of you.
I’m done filtering myself through the lens of “does this sound like an LLM?”
That filter was never about improving my writing. It was about performing a particular kind of humanness for an audience I assumed was suspicious. It was writing for an imaginary detector instead of an actual reader.
That’s not writing. That’s anxiety masquerading as craft.
The most authentic thing I can do is stop caring about this entirely.
Not in a defiant, performative way. Just in a genuine release of a concern that never served me.
I write how I write. Some of that might overlap with patterns that LLMs reproduce. Fine. Those patterns existed in human language before LLMs existed. They’ll exist after whatever comes next.
My voice is my voice. I don’t owe anyone proof that a human wrote this.
So yeah. I might use em dashes. I might write with structure. My transitions might flow naturally from one idea to the next.
And I don’t give a fuck if that sounds like AI.
Because it doesn’t. It sounds like me.