Typographic punctuation
EM DASH (U+2014)
The long dash, one em wide. It sets off a parenthetical clause, or marks an interruption where a comma would be too quiet.
- Codepoint
U+2014- Also written as
u2014\u2014—- Category
- Typographic punctuation
- Removed by default
- No
- Governing option
- Punctuation
--punct - Can be load-bearing
- No
Copy and paste U+2014
To copy and paste EM DASH (U+2014), press the button: the character itself goes to your clipboard, and Ctrl+V (Cmd+V on a Mac) pastes it anywhere you can type.
Where it comes from
Autocorrect in every word processor (two hyphens become an em dash), publishing house style, Markdown processors with "smart punctuation" on, and writers who type it deliberately.
Why it matters
It is not hidden and it is not a problem, but it is a different codepoint from a hyphen, so it breaks naive text comparison, and it is the single most cited "AI tell" on the internet, which makes it worth stating the facts about plainly.
Before and after
What `--punct` does to a spaced em dash. Note this is a typography change, not a cleaning: nothing was hidden.
The character is shown as a labelled chip so you can see where it sits. In your text it draws nothing at all.
When it is legitimate
Whenever you want an em dash. It is correct English punctuation with several centuries of history behind it.
How to remove it
Turn on Punctuation (`--punct`). A dash with text on both sides becomes " - " (spaced hyphen); other typographic dashes and minus signs become a plain hyphen.
Working in a file rather than a paste? The document report for .docx, .odt and .html finds it inside the file, together with hidden runs and tracked changes, without you having to open it. If the question is what the rest of the writing does rather than what this one character does, count the dashes and stock transitions and see where each habit sits, with no score and no claim about who wrote the text.
Is it an AI watermark?
Appears in some AI output, but is not a watermark
Models do use em dashes heavily, because they were trained on edited prose and edited prose uses them. That is a STYLE observation, not a watermark: it is not deliberate, not detectable-by-design, and not evidence about any single document. Plenty of human writers use more em dashes than any model. If you want to know whether your own draft leans on them, the prose-habits report counts them per thousand words and lets you decide. It will never tell you a text is machine-written.
It genuinely turns up in model output, as ordinary typography learned from the training data. It is not deliberate, not detectable by design, and not evidence about any single document.
Questions
Does a lot of em dashes mean AI wrote it?
No. It is a stylistic tendency shared by language models and by well-edited human prose. No character count can establish authorship, and any tool that gives you a percentage from one is inventing it.
Should I convert them to hyphens?
Only if you need plain ASCII: a code field, a legacy system, a plain-text email. In prose the em dash is correct punctuation and converting it makes the writing worse, not more human.
What exactly does `--punct` change?
A dash between two non-space characters becomes " - "; en dashes, horizontal bars, minus signs and the various hyphens all become "-"; curly quotes become straight; an ellipsis becomes three dots.
Check your own text
Paste anything into the cleaner to see every hidden character it contains, with position, codepoint and what happens to each one. Nothing is uploaded.
Check your own textRemove em dashes from ChatGPT text (and the characters you cannot see)Clean it in Chrome as you pasteClean it on your Mac with one keystrokeAll characters