Just as there are ways of speaking - accents, expressions, local idioms - there are ways of writing. We’re wired by evolution to imitate. Body language mirrors. Speech patterns converge. Most of it happens below conscious awareness.

In the military, an adjutant is the assistant to a colonel. They produce documents in a very specific form of business English - its own conventions, euphemisms, grammar. Adjutants aren’t linguists. They follow the conventions without realising, indoctrinated by years of exposure to military documents.

I suspect something similar is happening with LLMs.

The Tells

Anyone who uses the models all day knows the patterns. “It’s not X, it’s Y.” Em dashes everywhere. “You’re absolutely right.” Parallel sentence structures. Tidy summaries.

Most conversations about this have focused on detection - is this essay AI-written? Is this account a bot?

The underexplored question: are humans starting to write like AI?

I searched for typical LLM expressions in messages I could see - group chats, direct messages between colleagues, places where nobody would use a model to communicate.

They were there. More than I expected.

Models are trained on us, so they didn’t invent these patterns. But given how deeply wired imitation is, it would be strange if we weren’t picking up habits from our new interlocutors. We do it unconsciously with people. Why not with machines we talk to all day?

The Adjutant Again

The adjutant doesn’t know they’re writing in a house style. They’ve just read enough signals to know what a proper document looks like. The conventions feel natural because they’ve been absorbed through repetition.

I wonder how much of our prose is starting to feel natural for the same reason.