Em Dash Remover — Replace Em Dashes in AI Text
Detect every em dash and en dash in your text, see how many appear, and replace them with a comma, a space, or a regular hyphen — all in your browser.
Em dashes (—): 0 · En dashes (–): 0
What em dashes are and why AI writing is full of them
An em dash (—, U+2014) is a long horizontal stroke, wider than either a hyphen (-) or an en dash (–, U+2013). Its name comes from typesetting: it is traditionally the width of a capital M in the font being set. In edited prose, an em dash can set off a parenthetical phrase, introduce an abrupt change of direction, or replace a colon or comma when extra emphasis is wanted. Used sparingly, it is a legitimate punctuation choice. Used in every other sentence, it reads as a stylistic habit — and that habit has become a recognisable fingerprint of AI-generated text.
Large language models learned to write from a corpus dominated by edited text: journalism, published books, Wikipedia, technical documentation. All of those sources use em dashes at a certain frequency, and the models learned both that em dashes exist and something approximating when to use them. The problem is that they learned a stronger association than most human writers maintain. The result is AI output that deploys em dashes at a rate high enough that editors and publishers now flag it as a probable signal of synthetic origin, independent of any specific detection algorithm.
How to use this tool
Paste your text into the input box. The tool immediately counts how many em dashes (—) and en dashes (–) it found and shows the totals.
Choose your replacement: a comma followed by a space (the most semantically close substitute in most sentences), a single space (which removes the dash while preserving the gap between words), or a regular hyphen (for contexts where a shorter dash is acceptable). The en dash option lets you handle both dash widths together or leave en dashes in place.
Press Remove Em Dashes. The output box shows the revised text. Copy the result with the Copy button, or clear both fields with Clear.
Before and after: replacing em dashes in real sentences
Original: "The project — which had run for three years — finally delivered its first working prototype." With "comma + space": "The project, which had run for three years, finally delivered its first working prototype." The em dashes here flanked a parenthetical phrase, and commas do the same job.
Original: "He opened the door — and stopped." With "space only": "He opened the door and stopped." The dramatic pause the dash was marking becomes a simple conjunction.
Original: "The answer is simple — start earlier." With "comma + space": "The answer is simple, start earlier." This works, though strictly speaking the result is a comma splice. In informal writing that is often acceptable; in formal prose a colon or period is more correct. The tool does the substitution; you decide whether to edit further.
AI text often chains em dashes across adjacent sentences, producing a rhythm that reads as mechanical once you notice it. Processing the full document and scanning the output reveals that pattern more clearly than reading the original — which is also a useful diagnostic step even if you end up restoring some dashes.
Why em dashes became a signal of AI-generated text
Frequency is the primary issue. Independent analyses of output from several large language models consistently find em dashes appearing at roughly three to five times the rate found in matched sets of human-written articles on the same topics. Any single instance is stylistically defensible; the aggregate rate is statistically distinguishable from most human writers' habits.
Context independence is the second issue. Human writers tend to use em dashes for specific rhetorical effects: an abrupt interruption, a dramatic reveal, a parenthetical aside that doesn't warrant full parentheses. AI models distribute em dashes more evenly across sentence positions, producing a flatter pattern that doesn't track those rhetorical functions as reliably.
The pattern is now widely discussed. Writing coaches, content editors, and academic integrity staff have each published their own version of an em-dash heuristic. The heuristic is not foolproof — some human writers use em dashes heavily, and AI output can be revised to avoid them — but the pattern is consistent enough that its presence raises the probability of AI authorship in both automated and manual review.
Replacing em dashes does not make AI text human. The underlying sentence structures, hedging patterns, and topical scope remain. But it removes one of the more immediately visible surface markers, which is a legitimate editing goal: choosing not to use em dashes is a stylistic decision that every writer makes, and revising for style is normal.
Six situations where removing em dashes matters
Academic submission. Many style guides — APA, Chicago, and several journal-specific ones — permit em dashes but specify they should be used sparingly. A draft that contains twenty em dashes in eight pages will likely draw a comment from an editor regardless of how it was written.
Content marketing and copywriting. Some brand style guides explicitly ban em dashes as too informal or too disruptive to reading flow. A content team working partly with AI drafts needs a reliable way to sweep the output before publication.
Journalism and news writing. AP Style permits em dashes only in specific constructions. Other outlets have their own rules. Replacing em dashes and reviewing the substitutions is faster than hunting them manually across a long article, particularly one that was AI-drafted.
Publishing to systems that don't render em dashes correctly. Certain email clients, legacy CMS fields, and plain-text export pipelines display an em dash as a question mark, a box, or three hyphens, depending on their character encoding. Replacing them before export avoids the rendering problem entirely.
Reading your own AI usage patterns. Counting how many em dashes appear in a draft you thought was mostly human-written can be diagnostic. A high count may indicate more AI contribution than you remembered, or that the draft needs more thorough editing before it reads as your own voice.
Translating or adapting text. Translators working from AI-assisted source documents sometimes need to normalise punctuation before running the text through translation memory software. Translation memory segments on sentence boundaries, and an em dash mid-sentence can produce unexpected segment splits.
Four mistakes when editing em dashes by hand
Searching for the wrong character. An em dash (—) and an en dash (–) look similar at small sizes, and both are different from a hyphen (-) or two hyphens (--). A find-and-replace that searches for only one of them will miss the rest. This tool searches all three explicitly and shows each count separately.
Forgetting the surrounding spaces. Em dashes in AI output often appear without spaces around them — like this — but sometimes appear with a space on one or both sides. A find-and-replace with "comma" as the replacement that doesn't account for trailing spaces can produce commas followed by no space. This tool always produces a space after a comma replacement regardless of the spacing in the original.
Accepting the first substitution without reading it. A comma replacement is semantically close to an em dash in many constructions, but not all. "The rule is — never delete without a backup." becomes "The rule is, never delete without a backup." That comma feels wrong; you'd want a colon there instead. The tool does the mechanical work; a quick read of the output is still yours to do.
Replacing en dashes that should stay. En dashes serve a different purpose: they mark ranges — pages 14–22, March–June, New York–London. Replacing those with commas or spaces corrupts the meaning. If your text uses en dashes correctly for ranges, use the option to leave en dashes alone and only replace em dashes.
Your text never leaves your browser
All detection and replacement runs as JavaScript in the page you're already looking at. The text you paste is never sent to any server, never stored anywhere, and disappears when you close or refresh the tab. There is no account, no log, and no analytics on the content of your text.
If you're working with unpublished writing — a draft manuscript, a confidential report, a client document — browser-side processing means the content stays entirely on your machine.
Related tools on this site
If your text also contains invisible characters — zero-width spaces, byte-order marks, soft hyphens — the invisible character remover handles those in a separate pass.
For a full clean of AI text formatting residue, the clean AI text tool strips the invisible Unicode layer that AI interfaces add when you copy from a chat window.
If the text also has hard-wrapped line breaks from PDF export or email formatting, the remove line breaks tool on this site handles those. You can run both tools in sequence: remove line breaks first, then clean up the dashes.
The ChatGPT watermark remover page explains the broader range of surface-level patterns in AI text that are worth knowing about beyond em dashes.
Frequently Asked Questions
▸What is the difference between an em dash, an en dash, and a hyphen?
A hyphen (-) is the short stroke used for compound words and line-break hyphenation. An en dash (–, U+2013) is wider — roughly the width of a capital N — and marks ranges: 2010–2020, pages 45–60. An em dash (—, U+2014) is the widest and marks parenthetical asides, abrupt changes in direction, or strong emphasis. AI models use em dashes far more often than en dashes or hyphens for stylistic purposes, which is why em dashes are the main target here. The tool counts all three separately so you can see the full picture.
▸Why do AI language models use so many em dashes?
The short answer is that em dashes appear frequently in the high-quality edited text that made up the training corpus — journalism, essays, published books — and models learned to replicate that style. The longer answer is that an em dash is structurally flexible: it can introduce, interrupt, clarify, or emphasise without changing a sentence's grammatical structure. For a model generating tokens one at a time, it is a low-risk punctuation choice that works in many positions. The result is overuse at a rate most human writers don't sustain across a long document.
▸Should I replace every em dash with a comma?
Not automatically. Comma replacement works well when the em dash flanks a parenthetical phrase: 'The plan — if it works — will save weeks' becomes 'The plan, if it works, will save weeks.' It works less well when the em dash introduces a conclusion: 'There is one answer — start now.' becomes 'There is one answer, start now.' which is a comma splice. In those cases a colon or period is more correct. The tool does the substitution; reading the output and adjusting a handful of cases manually takes less time than finding every dash by hand.
▸What happens to en dashes — does this tool remove those too?
By default the tool replaces en dashes (–) using the same rule you chose for em dashes. If your text uses en dashes correctly for ranges — pages 40–55, 2019–2023 — you should review those positions in the output, or choose to leave en dashes alone. The tool shows the en dash count separately so you can see how many there are before deciding. If you have only a few and they're all range markers worth keeping, the simplest approach is to restore them manually after running the tool.
▸How many em dashes is too many in normal writing?
There is no hard rule, but a common guideline is one or two per page — roughly one per 300 to 500 words. At that rate, each em dash stands out as an intentional choice. Above that rate the repetition starts to register as a habit. AI-generated drafts routinely hit one per 50 to 100 words — five to ten times the guideline rate — which is why frequency alone is a useful diagnostic even before you read individual sentences.
▸Does removing em dashes affect my text's meaning?
In most cases, no — a comma or space substitute preserves the logical structure of the sentence. The exception is the appositive dash construction: 'She wanted one thing — quiet.' Replacing that with a comma gives 'She wanted one thing, quiet.' which is grammatically fine but loses some of the original emphasis. Whether that matters depends on whether the emphasis was intentional. In most AI-generated text it was not a deliberate rhetorical choice, so comma replacement is usually safe. When in doubt, read the output before copying.