Em Dash Replacer
The em dash is the most recognisable punctuation habit in AI writing, and the fastest thing to fix. Swap it for a comma, a hyphen or nothing at all, across the whole dash family, with every one marked in your text before it changes.
The punctuation mark that gives the game away.
There is nothing wrong with an em dash. It is correct punctuation, it has been correct for centuries, and good writers use it well. The problem is frequency. Assistants produce them at a rate well above ordinary human writing, because the published prose they learned from uses them heavily and because a dash is the easiest way to join two clauses without deciding exactly how they relate.
Readers noticed before the detectors did. An editor scanning a paragraph with three em dashes in it now forms an impression about where the text came from, and that impression arrives before any tool has scored anything. That is the real cost, and it is why this is worth fixing even though it is a small change.
The second reason is duller and more consequential. An em dash is not ASCII, so every system that assumes ASCII damages it, and always after publication rather than before.
Five reasons to take the dashes out.
- It reads as machine-writtenThis is the honest first reason. Assistants use em dashes at a rate well above most human writing, and readers have learned the pattern faster than any detector did. A paragraph with three of them in it reads a particular way to an editor now, regardless of what any tool would score it. Replacing them changes the impression the prose makes.
- A style guide says soAP style prefers a spaced hyphen. Many corporate and academic guides prohibit em dashes outright for consistency, and legal drafting avoids them because the meaning of an interrupted clause can be argued about. If you are writing to a guide, this is not a preference but a requirement.
- The destination mangles themEm dashes are not ASCII, so anything that assumes ASCII will damage them. Older email clients render them as question marks or boxes, SMS gateways substitute or drop them, and a CMS with the wrong encoding turns one into a run of three unrelated characters. The failure is always visible and always after publication.
- The text is going somewhere plainConfiguration files, code comments, command-line output, CSV exports and version-control diffs are all better off in plain ASCII. A dash that renders correctly in a document can produce an encoding error in a build pipeline.
- The writing leans on themIndependent of anything to do with AI, a dash is an easy way to avoid deciding what the relationship between two clauses actually is. Replacing them with commas, full stops or brackets forces that decision, and the writing is usually better for it.
Seven characters, and only one of them is on your keyboard.
Most of the confusion here is not knowing which dash you actually have. They look similar, they are different code points, and no spell checker distinguishes them.
- Em dash
U+2014The long one, as wide as a capital M. It sets off a phrase the way brackets or commas would, but with more force. This is the one AI writing reaches for constantly, and the one people mean when they say a piece of text reads like it was generated. - En dash
U+2013Shorter, as wide as a capital N. It belongs in ranges and spans: page numbers, date ranges, scores. Word converts a hyphen into one automatically in some contexts, which is why it turns up in text nobody deliberately typed it into. - Hyphen
U+002DThe one on the keyboard. It joins compound words and breaks words across lines. It is plain ASCII, which is why it is the safe replacement for everything else on this list. - Minus sign
U+2212A mathematical operator that looks almost exactly like an en dash and is a different character entirely. It arrives with figures copied out of spreadsheets and typeset documents, and it breaks numeric parsing quietly. - Horizontal bar
U+2015Used to introduce quoted dialogue in some languages. In English text it is almost always an artefact of a conversion, and it renders as a long dash that no keyboard produced. - Two-em dash
U+2E3ATwice the width of an em dash, used to mark omitted letters or words. Rare in writing and not rare at all in text that has been through a formatter. - Three-em dash
U+2E3BThree times the width, used in bibliographies to stand in for a repeated author name. If one appears in your prose, something in the pipeline put it there.
Choosing the replacement.
Replace with a comma
The default here, and the right answer for most prose. An em dash setting off a parenthetical phrase is doing a comma's job with more emphasis, so a comma reads naturally in its place: "the results were surprising—far better than expected" becomes "the results were surprising, far better than expected". The sentence survives intact and stops announcing itself.
Replace with a hyphen
Pure ASCII and maximum compatibility, which is what you want for technical destinations, plain-text formats and anything passing through a system you do not control. It reads slightly oddly in prose, because a hyphen is not really a substitute for a dash, but it will never render as a question mark.
Replace with a spaced hyphen
A hyphen with a space either side, which is the AP style convention and the closest ASCII approximation to how an em dash actually reads. Use this when the text is going somewhere plain but still has to read as writing rather than as data.
Remove entirely
Deletes the dash and leaves the words either side touching. This is occasionally what you want and usually not, since it welds two words together. Turn tidy spacing on if you use it, and read the result before you ship it.
Tidy spacing
Text that writes dashes with spaces around them—like this—leaves a double space behind when the dash becomes a comma, and a space sitting in front of that comma besides. This option closes both, so the replacement looks like something a person typed.
What this will and will not do to a detector score.
A good deal of the advice about em dashes is sold on the promise that removing them defeats AI detection. It does not, and it is worth saying so plainly rather than taking the sale.
Detection models score how predictable a text is: sentence-length variation, word choice, how closely each phrase follows from the last. Punctuation is one small input among many. Replace every dash in a document and the score moves a little, because you have changed a few characters out of thousands, and the phrasing that the model is actually measuring is exactly as it was.
What changes is the human read. A paragraph without three dashes in it does not announce itself, and it is a person who decides whether your draft looks machine-written long before any tool is opened. That is a better reason to do this than a score, and it is the one that holds up.
If invisible characters rather than punctuation are what you are chasing, the zero-width space remover covers those, and the per-model tools for ChatGPT, Claude, Gemini and Grok go further again. Spacing is a separate job handled by the AI space remover, and the ChatGPT text cleaner on the homepage runs every pass at once when a draft needs more than one.
Em dash questions.
What is an em dash?
A punctuation mark, U+2014, as wide as a capital M. It sets off a parenthetical phrase or marks an abrupt change of direction in a sentence, doing the work of brackets or a pair of commas with more emphasis. It is longer than an en dash (–) and much longer than a hyphen (-).
Why does AI writing use so many em dashes?
Because the writing it was trained on uses them, and because a dash is a low-risk way to join two clauses whose exact relationship has not been decided. Published prose, which is heavily represented in training data, uses em dashes far more than everyday writing does. The result is output that reads more like an essay than like an email.
Do em dashes make text look AI-generated?
To readers, yes, and increasingly so. To detectors, less than people assume: they score sentence rhythm, word choice and predictability rather than counting punctuation. Replacing every dash will not move a detector score much on its own. It will change how the paragraph reads to an editor, which is usually the thing that actually matters.
Will replacing em dashes beat an AI detector?
No, and it is worth being clear about it rather than selling you the opposite. Detection models score phrasing and predictability across whole sentences. Punctuation is a small input. Replace dashes because the text reads better and travels better, not because it changes a score.
What should I replace an em dash with?
A comma, for most prose: an em dash is usually doing a comma's job with more force, so the sentence reads naturally without it. Use a plain hyphen when the destination needs ASCII, and a spaced hyphen when you are writing to AP style or want the closest approximation to how a dash reads.
What is the difference between an em dash and an en dash?
Width and purpose. An em dash (—, U+2014) sets off a phrase inside a sentence. An en dash (–, U+2013) joins a range: pages 10–20, the 2024–2025 season. They are different characters, and a spell checker will not tell you which one you have.
Why do em dashes turn into question marks in email?
An encoding mismatch. The em dash is not part of ASCII, so a client or gateway that assumes ASCII substitutes a placeholder for it. The same cause produces boxes, or a run of three unrelated characters where one dash used to be. Replacing dashes with hyphens before sending removes the possibility entirely.
Does this remove en dashes too?
Only if you turn that option on, and it is off by default deliberately. En dashes are usually correct where they appear, since a number range genuinely wants one. Turn it on when you need pure ASCII output and are prepared for ranges to read as hyphens.
What are two-em and three-em dashes?
U+2E3A and U+2E3B, at twice and three times the width of an em dash. The first marks omitted letters or words, the second stands in for a repeated author name in a bibliography. Neither belongs in ordinary prose, and both turn up in text that has passed through a formatter.
Will this change my hyphens?
No. Ordinary hyphens are ASCII and are left exactly as they are, in compound words, phone numbers and everywhere else. Only the Unicode dash characters are touched.
Does removing dashes break my formatting?
No. Line breaks, paragraphs, spacing and every other character are preserved. The only change beyond the dash itself is the optional spacing tidy-up, which closes the gap a spaced dash leaves behind.
Can I replace em dashes in Microsoft Word instead?
Yes. Find and Replace accepts ^+ for an em dash and ^= for an en dash in the find box, which is quicker than pasting one in. The catch is that Word's autocorrect will convert your replacements back into dashes as you continue typing, so turn that off in the autocorrect settings first or you will be doing it twice.
How do I stop Word turning hyphens into dashes?
Turn off the em dash substitution in AutoCorrect Options, under AutoFormat As You Type. It is on by default, which is why documents accumulate dashes nobody typed. Google Docs has the same setting under automatic substitution.
Is my text uploaded anywhere?
No. Everything runs in your browser. Nothing is transmitted or stored, and closing the tab discards it.
Is it free?
Yes, with no limits and no account. It runs on your device, so there is nothing on our side to charge for.