The Ghost-in-the-Machine Problem With AI Writing

Readers who encounter generic AI content often can't name exactly what bothers them. The grammar is correct. The argument is coherent. But something feels off - a kind of hollowness, an absence of conviction. They scroll past without quite knowing why.

What they're detecting is the ghost-in-the-machine problem: technically correct content that has no one in particular behind it. A real person holds opinions, makes unexpected word choices, takes risks in how they state things. A default model regresses to safe, averaged expression.

Presence in writing is not a stylistic adjective. It's the accumulated effect of specificity: a real example with real details, a position that could be argued against, a moment where the writer says something that not everyone would say. These are signals that a real person, with real stakes in the argument, wrote this.

You can't manufacture presence by describing it. You have to write in a way that expresses it - which is why grounding AI drafts in your actual writing corpus produces presence, while prompting a model to "sound more human" does not.

The instinct when AI writing feels hollow is to add style prompts: "be more conversational," "add personality," "sound less formal." These help at the margin. The real fix is grounding: give the model your real posts so it drafts from your specifics rather than its averages. What comes out still needs your edit - especially the opinion the model avoided - but it starts from presence rather than working toward it.