Your AI Writing Has a Tell, and It's the Em-Dash

AI text has fingerprints, and the em-dash is the most obvious. String three of them in a row and any reader who's seen generated prose knows. It's not a rule, it's a tell.

Other tells: 'it's worth noting,' 'in today's fast-paced world,' the list that pretends to be insight, the conclusion that concludes nothing. They're habits models fall into because they're statistically safe.

I killed them by editing the draft as if a human wrote it badly, not as if a machine wrote it perfectly. A person varies sentence length, uses fragments, sometimes sounds unsure. The model avoids all of that.

The goal isn't to hide AI, it's to sound like you. Readers forgive imperfect human writing; they resent fluent generic writing. The tell is the gap between the two.

If your writing feels off, read it for the tells, not the typos. The dashes and the filler are what give you away, and they're easy to cut.