Dashes: U+2014, U+2013 and the rest

The dash family: em dash, en dash, figure dash, horizontal bar, and the plain Unicode hyphen. All are visible, all are legitimate punctuation, and all read as a dash of some length that an ASCII hyphen approximates.

How it works

The dash family is a set of distinct code points that all render as a horizontal stroke of different lengths: the hyphen-minus at 0x2D that your keyboard produces, U+2010 the true hyphen, U+2011 its non-breaking twin, U+2012 the figure dash sized to align with digits, U+2013 the en dash for ranges, U+2014 the em dash for parenthetical breaks, and U+2015 the horizontal bar for quotations. Word processors and publishing systems substitute the typographic ones automatically, and language models emit them because their training text is full of professionally typeset prose. That last fact is why this group matters for detection rather than for correctness. Detectors count characteristics that separate machine text from typed text, and em dash frequency is among the most reliable single features available, because a keyboard has no em dash key and most people never learn the shortcut, while a model reproduces the distribution of edited prose. The technical failures are ordinary: a dash that is not a hyphen-minus breaks a command line, a configuration key, a CSV field or a lookup.

What it is legitimately for

Real typography. An em dash sets off a clause, an en dash spans a range, a figure dash groups digits. Nothing is wrong with any of them; they are simply not what a keyboard produces.

What it breaks

• Nothing technical, usually. This is the group that costs you in a different way.
• Em dash frequency is the single most reliable stylometric signal detectors use, because models emit them far more often than people type them.
• In code, a configuration file or a command line, a dash that is not a hyphen produces an error that reads as nonsense.

How to find it yourself

You can see them. If a passage has more em dashes than you would ever type, that is the tell, and it is the one a reader notices too.

What Clipboard Sanitizer does

Replaced with an ASCII hyphen. This changes text you can see, so it is the rule most people turn off, and turning it off is a perfectly reasonable choice.