Why your AI draft sounds fake, and the 60-second fix.

Four reasons, all from how these models write.

1. It's the average of everything

A language model predicts the most likely next word. Likely means average. Every sentence trends toward the centre of how everyone writes, which is how nobody writes. Specificity is rare by definition, and rare is what reads as real.

2. Scaffolding instead of substance

Stock openers ("In today's fast-paced world"), stock joins ("Furthermore,"), stock bows ("In conclusion"). The structure of an essay with none of the stakes. Readers feel the template before they finish line one.

3. Enthusiasm with no evidence

Thrilled, humbled, game-changer, transformative: high-wattage words attached to ordinary news. Humans reserve those words for moments; models dispense them by default. The gap between the claim and the event is exactly what "fake" means here.

4. Symmetry and summary

Even paragraphs, even rhythm, and a closing that restates the opening. School-essay shape. Human posts are lopsided: one long thought, one short jab, a question left open.

The 60-second fix

1. Strip the scaffolding (10s). Paste the draft into the cleaner. It removes the stock openers, joins and sign-offs, and shows each one as a chip so you can see what was machine.
2. Add one detail only you know (30s). A number, a name, a date, a thing someone said. One is enough. This is the part no model can write, and it's the part that makes the post yours.
3. Rewrite the last line (20s). Not a summary. A question you'd actually like answered, or one flat sentence of opinion.

That's it. AI as the first-draft machine, you as the editor with the specifics. It's the strongest combination available right now, and it reads like a person because, at the points that matter, one was.