How to Remove Markdown, Asterisks and Em Dashes from ChatGPT Text

You ask ChatGPT for a paragraph, copy it into an email or a Google Doc, and it arrives looking like this: **Key point:** the results — mostly positive — were ## strong. The words are fine. Everything around them is noise. This guide explains where that noise comes from, what exactly gets left behind, and four ways to get rid of it, from a one-click fix to a find-and-replace you can run in Word.

Why AI text pastes with asterisks and hash signs

ChatGPT, Claude, Gemini and Copilot all write their answers in markdown, a simple formatting language invented in 2004 for writing on the web. In markdown, **bold** means bold, *italic* means italic, a line starting with ## is a heading and a line starting with - is a bullet. The chat window reads those symbols and shows you nicely formatted text.

What happens next depends on how you copy. The copy button in most chat apps puts the raw markdown on your clipboard, symbols included. Selecting the text with your mouse usually copies the formatted version instead, but that brings its own problems: heading sizes, fonts and background colours that fight with your document. Apps that do not understand markdown, such as Gmail, Outlook, LinkedIn, Excel and most CMS text fields, then show the symbols as plain characters.

Advertisement

What gets left behind

These are the leftovers that cause most of the trouble:

You see	What it was	What you want
**word**	Bold	word
*word* or _word_	Italic	word
## Title	Heading	Title on its own line
- item or * item	Bullet	A bullet your app understands
[text](https://…)	Link	The text, the URL, or both
| A | B | and |---|	Table	Rows you can paste into a spreadsheet
`code`	Inline code	code
—	Em dash	A comma, a hyphen or a new sentence
[1], 【4:0†source】	Citation markers	Nothing

There is also a group you cannot see: hidden characters. Zero-width spaces, non-breaking spaces and byte-order marks can ride along when text moves between apps. They are invisible, but they break things. A word containing a zero-width space will not match a search or a spell check, a “blank” cell in a spreadsheet is not really blank, and code containing a non-breaking space may refuse to run.

Method 1: clean it in one click

The fastest route is a cleaner built for this job. Our ChatGPT Text Cleaner removes all of the leftovers above in one pass:

1. Paste the AI text into the left box.
2. Choose where it is going: plain text, email and documents, or a social post.
3. Copy the clean version from the right-hand box.

It shows exactly what it changed (“22 markdown marks removed, 2 em dashes replaced, 1 hidden character removed”), keeps code blocks untouched, and leaves things that only look like markdown alone, such as 2*3, #hashtags and file_names_with_underscores. Everything runs in your browser, so nothing you paste is uploaded.

Method 2: ask for plain text in the first place

If you know the text is going somewhere plain, tell the assistant before it writes. A line like this at the end of your prompt works well:

Write in plain text with no markdown: no asterisks, no hash signs, no bullet symbols, no em dashes. Use short paragraphs.

This helps, but it is not reliable. Assistants drift back to their habits in long answers and in follow-up messages, and the instruction does nothing for text you already have. It also cannot stop hidden characters that appear during copying. Treat it as a first filter rather than a fix.

Method 3: find and replace in Word or Google Docs

For a one-off document you can clean the most common symbols with find and replace.

Google Docs: open Edit → Find and replace, tick Use regular expressions, and run these one at a time, replacing with nothing unless noted:

• \*\* removes double asterisks (bold).
• ^#+\s removes heading hash signs at the start of lines.
• — finds em dashes; replace with a comma and a space, or a spaced hyphen.

Microsoft Word: open Replace (Ctrl + H), type ** in Find and leave Replace empty, then Replace All. For em dashes, Word’s special character list under More → Special includes the em dash. Word’s wildcard mode can do more, but its syntax differs from standard regular expressions, so test on a copy first.

The weakness of this method is coverage. Each symbol needs its own pass, single asterisks for italics are hard to target without damaging maths, links and tables need manual work, and you will not see hidden characters at all.

Method 4: regular expressions for developers

If you clean AI output inside your own code, a handful of patterns cover most cases. In JavaScript:

text
  .replace(/\*\*(.+?)\*\*/g, '$1')          // bold
  .replace(/^#{1,6}\s+/gm, '')               // headings
  .replace(/\[([^\]]+)\]\([^)]+\)/g, '$1')   // links: keep the text
  .replace(/`([^`]+)`/g, '$1')               // inline code
  .replace(/[\u200B-\u200D\uFEFF]/g, '')     // zero-width characters
  .replace(/\u00A0/g, ' ')                   // non-breaking spaces
  .replace(/\s*—\s*/g, ', ');                // em dashes

Be careful with the simple versions. A naive italic rule such as \*(.+?)\* will eat the asterisk in 2*3*4, and a naive underscore rule breaks snake_case names. Code blocks also need to be protected before you touch anything, or the cleaner will strip symbols that the code needs.

Should you remove em dashes?

The em dash is a perfectly good punctuation mark, and plenty of human writers use it. It has become a talking point because AI assistants use it heavily, often several times per paragraph, and some readers now treat a page full of dashes as a sign of unedited AI text. If your goal is text that reads naturally, the fix is to vary the punctuation, not to ban it: some dashes become commas, some become full stops, and an occasional one stays. The cleaner’s “Split into sentences” option is a good starting point when a paragraph has too many.

Be realistic about what cleaning can do. Removing symbols makes text tidy; it does not change the wording, structure or rhythm that make writing sound generated. If the writing matters, read it through and rewrite it in your own voice.

Quick settings for common destinations

• Gmail and Outlook: the Email & docs preset. It keeps curly quotes, uses • bullets and keeps link addresses in brackets.
• Word and Google Docs: Email & docs as well. Paste with Ctrl + Shift + V (Cmd + Shift + V on Mac) to drop fonts and colours.
• LinkedIn, X and Instagram: the Social post preset. It keeps emojis, uses • bullets and shows bare URLs.
• Excel and Google Sheets: Plain text with Tables set to “Tabs”, so each table cell lands in its own spreadsheet cell.
• WordPress and other CMSs: Plain text, then add headings with the editor’s own tools so they are real headings rather than bold lines.
• Code, CSV and JSON: Plain text with curly quotes turned off in the output, so every quote is a straight quote the parser understands.

Frequently asked questions

Why does copying from ChatGPT sometimes keep the formatting and sometimes not?

The copy button copies raw markdown. Selecting text with the mouse copies the rendered, formatted version. Which one arrives cleanly depends on the app you paste into.

Are the hidden characters in ChatGPT text a watermark?

Not in any way the AI companies have announced. Zero-width and non-breaking spaces mostly appear when text passes between apps. They are still worth removing because they break searches, word counts and code, and the cleaner reports how many it found.

Will cleaning the text make it pass an AI detector?

No. Detectors look at wording and sentence patterns, not at asterisks. Cleaning fixes presentation only.

Is it safe to paste confidential text into an online cleaner?

Only if the cleaning happens on your device. Our cleaner runs entirely in your browser and never sends your text anywhere. For other tools, check whether they upload your text before using them with anything sensitive.

Try it now: paste your next ChatGPT answer into the ChatGPT Text Cleaner and copy clean text in one click.