Your Boss Says Your Writing Sounds Like AI. They're Not Wrong.

Your director forwards the doc back with a one-line note: "this sounds like AI." You did use ChatGPT for the first draft, then rewrote half of it by hand, and the comment still lands hard, because your name is on the file and now someone thinks you phoned it in.

You're not doing anything wrong by drafting with a model. Most people at your company are, whether they admit it or not. The problem isn't that you used AI. It's that the AI's fingerprints survived your edit.

The median is the default, and the default is the tell

Ask a model to write a professional update and it reaches for the middle of everything it has seen. Sentences land at similar lengths. Paragraphs expand and wind down on schedule. Word choices never risk sounding strange or specific. That average is competent and completely anonymous. It reads like a company instead of a person, and your reader clocks it in about two sentences.

When you rewrite that draft, you probably swap a few phrases and tighten a sentence. That fixes vocabulary and leaves the architecture untouched, and the architecture is what gives the draft away. The rhythm stays even. The closing line restates the point one size larger than the one before it. A reader feels that shape even when no single word is wrong.

"Sounds like AI" is a credibility problem now

A year ago that note meant your prose was a little stiff. It carries a sharper charge today: it suggests you didn't think hard about the thing you sent. In a lot of offices, "sounds like AI" has quietly become shorthand for "didn't bother." Fair or not, that judgment attaches to you, not to the model you borrowed.

That's the part worth taking seriously. The stiff writing is a symptom. The reputation cost is the actual problem, and it compounds every time your name sits on top of copy that reads like it wrote itself.

Humanizers move the problem, they don't solve it

The tools that promise to humanize AI text mostly shuffle synonyms and chop up a few sentences. They treat a structural problem as a vocabulary one. The flat rhythm survives, now with odd word choices layered over it, and the writing reads generic and slightly off at the same time. You added a step and kept the tell.

Editing after the fact runs into the same wall. The pattern lives in the shape of the piece, not in any one line, so cleaning the lines leaves the shape in place.

Fix the draft before it's a draft

The pattern is easier to prevent than to sand off later. Shape how the model writes before it starts, and the first draft already varies its rhythm and structure, so your edit becomes a real edit instead of a rescue.

That's what Prose Coach does. It loads a set of writing rules into the model's prompt, so ChatGPT or Claude drafts with the variation built in and leaves room for your voice to show up. It won't supply the ideas, and it won't turn a thin update into a sharp one. What it strips out is the anonymous default, the part that made your director look twice.

The goal was never to hide that you use AI. Everyone uses AI. The goal is for the writing to sound like it came from someone who had a point, because you did. Get that back and the note stops coming.