Generic tells are the phrases and shapes that mark writing as unedited AI output: the same three openers, headings phrased as questions, a rash of em dashes, tidy emoji bullet lists. None is wrong alone. Together they are why so much AI text reads as nobody's.
Why it matters
AI writing has a house style, and readers have learned to spot it. The moment a post reads like the default model, it also reads as low-effort, and the brand behind it looks like it did not bother.
The tells are structural, not just word choice: how pieces open, how headings are phrased, how lists are formatted, how often the same sentence shape repeats. A human editor fixes these without thinking; unedited output leaves them all in.
Catching them is a cleanliness check, not a taste judgment. It says this reads as machine-made, here is where, before it ships under your name.
How Jinn treats it
Jinn runs a deterministic scan against a fixed catalog of the shapes that give AI text away and scores a draft for cleanliness, where higher means cleaner. It masks quoted and code passages first, so a legitimate quote is never flagged.
It fails safe: an empty or mostly-quoted draft comes back as unavailable, never as a false clean pass. On top of the generic catalog, Ghost also learns the phrases a specific voice has actually struck out, so the tells it flags become yours over time, evidenced by your own edits.
The tells that make writing read as nobody's.
Machine-made is a shape you can catch before it ships. See how Ghost keeps writing sounding like you.