The Em Dash Tell: Why ChatGPT Overuses It
What actually causes ChatGPT's habit of leaning on the em dash, what job the mark is doing in a sentence, and what a human editor reaches for instead: a comma, a colon, or a full stop.
Page 30 of 33.
What actually causes ChatGPT's habit of leaning on the em dash, what job the mark is doing in a sentence, and what a human editor reaches for instead: a comma, a colon, or a full stop.
A working reference for the vocabulary that marks machine-drafted prose, whichever assistant produced it: more than forty overused words organised by register, fifty stock phrases mapped to the one word that replaces each, the research on why models favour this exact list, and the editing pass that clears it out of a finished manuscript.
Getting caught is rarely a single moment triggered by a percentage score. It is a defined process with stages, evidence standards and appeal rights, most of which never make it past a first conversation. Here is what that process actually looks like end to end, using a published university procedure rather than a guess.
Detection software matters less here than most students assume. This walks through what a professor who has already read your work actually notices when a draft stops sounding like you, from a landmark blind study of examiners to the specific, checkable signs that show up long before anyone runs a scan.
A 2023 Stanford study found AI detectors misreading fluent non-native English writing as machine-generated at a much higher rate than native writing. Here is the mechanism behind that gap, and what happened when the same essays were rewritten with richer vocabulary.
Nearly every page ranking for this question is sold by a company with a humanizer to sell. Here is the mechanism instead: what these tools actually change, why some of that moves a detector's score and some does not, and what an aggressive rewrite risks in meaning and citations.