Guide · By PostBoost team · · Markdown
To remove em dashes from ChatGPT text, do two things. Stop new ones: put a no-em-dash rule in ChatGPT's custom instructions, which it has followed since November 2025. Then clean the ones you already have, replacing each dash by context: a comma between two words, a hyphen in a number range or after punctuation. Find and replace in a doc swaps every dash for the same mark. The free AI text cleaner applies the context rules for you and also fixes curly quotes, hidden characters and stray markdown.
Why bother? Readers now treat a caption full of long dashes as a sign it was written by AI. Wikipedia's dash article cites a 2025 Rolling Stone report on the nickname "ChatGPT hyphen", and an OpenAI spokesperson answered that ChatGPT's style depends on the prompt and is no proof of machine authorship. The mark itself is fine. The problem is that people stop reading.
The em dash, the en dash and the hyphen
Three characters look alike and do different jobs:
| Character | Looks like | Unicode | Normal use |
|---|---|---|---|
| Em dash | — | U+2014 | A break in a sentence: "It works — mostly." |
| En dash | – | U+2013 | Ranges: "pages 38–55", "2020–2024" |
| Hyphen | - | U+002D | Compound words: "well-known", the key on your keyboard |
AI writers use both dashes. A good fix treats them differently: the en dash in "2020–2024" should become a hyphen, not a comma.
1. Stop ChatGPT adding em dashes (custom instructions)
On 14 November 2025 Sam Altman posted that if you tell ChatGPT not to use em dashes in your custom instructions, "it finally does what it's supposed to do" (TechCrunch). Two limits matter. The rule has to be in custom instructions, not only in a chat message. And the default did not change: ChatGPT still writes em dashes for anyone who has not set the rule.
To set it:
- In ChatGPT, open Settings and then Personalization (OpenAI help).
- Open Custom instructions and make sure customization is switched on.
- Add a line like this one:
Never use em dashes (—) or en dashes (–). Use a comma, a full stop or parentheses instead. For number ranges, use a hyphen (2020-2024).
The instruction applies to your account only. Text from a colleague's ChatGPT, from Claude, Gemini or a tool that writes for you still arrives with dashes, so cleaning the final text stays the reliable step.
2. Find and replace in your document
If the text is already in a document, a find and replace is the quickest manual fix:
- Google Docs: Edit → Find and replace, paste — into Find, type your replacement, then Replace all. Repeat with – for en dashes.
- Microsoft Word: open Replace, type
^+in Find what for an em dash or^=for an en dash (Dummies).
The catch: one replacement for every dash. Replace them all with a comma and "2020–2024" becomes "2020, 2024", and "Results? — mixed" becomes "Results?, mixed". Replace them all with a hyphen and "week—and" becomes "week-and", one glued word. Read the result line by line.
3. Clean the text with rules that read the context
The AI text cleaner runs in your browser, and your text stays on your device until you click Schedule this, which opens the cleaned text as a draft in PostBoost. Paste the text, choose what Em dashes to becomes (a comma, a spaced hyphen ( - ), a hyphen (-), or keep dashes), and copy the result. With the default, a comma, it applies three rules in this order:
- An en dash between two digits becomes a hyphen:
2020–2024→2020-2024. - A dash between two words, with or without spaces around it, becomes a comma and a space:
a week—and it shows→a week, and it shows. - Any other dash, after a full stop, question mark or other punctuation, or at the start of a line, becomes a hyphen:
Results? — mixed→Results? - mixed.
Here is a real run. This text went in:
Scheduling isn’t the hard part — consistency is.
We tested it from 2020–2024. Results? — mixed…
**Let’s dive in.**
And this came out, with the default settings:
Scheduling isn't the hard part, consistency is.
We tested it from 2020-2024. Results? - mixed...
Let's dive in.
The summary under the result counted 3 dashes, 2 curly quotes, 1 ellipsis and 1 markdown marker, and it listed "let's dive in" as a stock AI phrase. It did not rewrite that phrase. Phrases like delve, game-changer or it's important to note are flagged for you to change in your own words; the cleaner only touches punctuation, spacing and formatting marks.
The other switches each have their own checkbox:
- Curly quotes to straight: ’ and “ ” become ' and ".
- Invisible characters: zero-width spaces, word joiners and soft hyphens are removed; non-breaking and thin spaces become normal spaces. The joiners inside emoji stay, so family and flag emoji are not split.
- Double and trailing spaces: collapses double spaces between words and removes spaces at line ends.
- … to ...: the single ellipsis character becomes three dots.
- Markdown
**and_markers: removes the asterisks and underscores chat tools use for bold and italic, which social networks show as plain characters. It leaves snake_case names, @handles and links alone. To keep the bold instead, use the LinkedIn post formatter (why it uses Unicode letters).
Check what a tool cannot decide
A rule picks a mark; it cannot read your meaning. After any clean-up, reread three cases:
- Two sentences joined by a dash. "Scheduling isn't the hard part, consistency is." is readable, but a full stop is stronger: "Scheduling isn't the hard part. Consistency is."
- A range written with an em dash. Only the en dash between digits counts as a range. If the AI wrote
2020—2024with an em dash, the comma setting turns it into2020, 2024. Fix that one by hand. - Paired dashes around an aside. "The tool — free for now — is fast" becomes "The tool, free for now, is fast", which works. If the aside is long, parentheses may read better.
Then check the length with the character counter: every network counts differently, and on X a link always counts as 23 (X character limit).
Clean text with the API or an AI agent
If your posts come from a script or an agent, run the same rules there. The cleaner is a free endpoint, no token needed:
curl -X POST https://postboost.co/app/api/tools/ai-text-cleaner \
-H "Content-Type: application/json" \
-d '{"text": "Most teams post 3–5 times a week—and it shows.", "dashes": "comma"}'
The response holds the clean text, one count per type of change, and any flagged phrases with their position:
{"text": "Most teams post 3-5 times a week, and it shows.", "changes": [{"type": "dashes", "count": 2}], "flags": []}
dashes takes comma, hyphen, spaced or keep; quotes, invisible, spaces, ellipsis, markdown and flag_phrases switch the other clean-ups on or off (all on by default). For an AI assistant, connect the free tools MCP server at https://postboost.co/mcp/tools (setup) and ask it to clean a draft with the clean-text tool; on the main PostBoost MCP server the same tool is cleanText. Fields, limits and SDK examples are in the AI text cleaner docs.
Fewer dashes from the start in PostBoost
When PostBoost writes captions for you, from a blog post or with the repurpose tool, it follows your workspace's Brand Voice: open it in the workspace sidebar and list stock phrases under words to avoid, next to example posts that show how you really punctuate. The brand voice guide walks through each field, and AI features covers captions and images. Before you schedule, paste the final text into the cleaner, then into the post editor, which checks each network's limits as you type.
Do this in PostBoost
Free tools for this guide
Free tool
AI text cleaner: remove em dashes, curly quotes and hidden characters
Remove em dashes, curly quotes, invisible characters and markdown from AI text, and spot phrases like delve. Free, runs in your browser.
Free tool
LinkedIn post formatter with bold, italic and a feed preview
Format LinkedIn posts with bold, italic, bullet and numbered lists, keep line breaks and preview the see more fold. Free, runs in your browser.
Free tool
Social media character counter
Free character counter for X, Threads, Bluesky, Instagram, LinkedIn, TikTok and more. Counts as each network does, live in your browser.
FAQ
How do I stop ChatGPT from using em dashes?
Add a rule to your custom instructions (Settings, Personalization, Custom instructions), for example: Never use em dashes or en dashes; use a comma, a full stop or parentheses instead. OpenAI said in November 2025 that ChatGPT now follows this rule when it is set there. It does not change the default, so other people's ChatGPT and other AI tools still add them.
Why does ChatGPT use so many em dashes?
OpenAI has not published a reason. The em dash is ordinary punctuation that writers used long before AI, and an OpenAI spokesperson has said ChatGPT's style depends on the prompt. Readers now treat a post full of em dashes as a sign it was machine-written, which is why people remove them.
What should replace an em dash?
It depends on the sentence. Between words, a comma usually works for an aside. In a number range such as 2020–2024, use a hyphen. After a full stop or question mark, or at the start of a line, use a hyphen so you never get a comma after punctuation. If the dash joins two full sentences, a full stop reads better than any dash or comma.
Can I find and replace em dashes in Word or Google Docs?
Yes. In Google Docs, open Edit, then Find and replace, and paste an em dash into the Find box. In Microsoft Word, type ^+ in Find what for an em dash and ^= for an en dash. Both replace every dash with the same thing, so check the sentences that need a different mark.
Does removing em dashes make AI text undetectable?
No. Em dashes are one habit among many. What makes a post read as human is what it says: your own example, a real number, an opinion. Clean the punctuation, then rewrite stock phrases such as delve or game-changer in your own words.
Is there an API to remove em dashes?
Yes. PostBoost's AI text cleaner is also POST https://postboost.co/app/api/tools/ai-text-cleaner, free and with no token needed, and the clean-text tool on the free tools MCP server at https://postboost.co/mcp/tools. It returns the cleaned text, a count of each change and the stock phrases it found.