You asked ChatGPT for a paragraph, copied the answer, and pasted it into Word. Now the document is full of ** around every bold phrase, curly quotes that don't match the rest of the file, and dashes twice as long as normal. Sometimes the spacing looks broken for no obvious reason.
This guide covers what that formatting is, why it survives the paste, and two ways to get rid of it.
What ChatGPT formatting actually is
ChatGPT writes its answers in Markdown, a plain-text markup convention. Inside the chat window the interface renders it nicely: bold looks bold, headings look like headings. When you copy the answer, the clipboard often carries the raw source, markers included:
**bold**,*italic*, backticks and##heading markers show up as literal characters- Straight quotes become typographic "smart quotes" and apostrophes
- Hyphens become em dashes (—) or en dashes (–)
- Three periods collapse into the single ellipsis character …
- Regular spaces become non-breaking spaces, and invisible characters like the zero-width space come along for the ride
None of this is visible in the chat window. All of it lands in your document.
For reference, the full set the cleaner catches, by code point:
U+200Bzero-width space — splits words in two for search and spell-checkU+00A0non-breaking space — looks like a space, breaks number parsing and string comparisonsU+FEFFbyte order mark — breaks JSON parsing and template rendering mid-fileU+200E/U+200Fdirectional marks — scramble the order of numbers and punctuationU+00ADsoft hyphen — invisible until a line wraps in the wrong spotU+2014/U+2013em and en dashes — the long dashes AI models loveU+2018–U+201Dsmart quotes — clash with straight-quoted documentsU+2026horizontal ellipsis — one character standing in for three dots
Why pasting into Word or Google Docs keeps the junk
"Paste without formatting" sounds like the fix, but it isn't. The Markdown markers are not formatting applied to the text; they are characters inside the text itself. Google Docs' "clear formatting" and Word's "keep text only" paste leave ** and ## exactly where they were. Email clients and CMS editors behave the same way.
The invisible characters are worse. You can't see them, so you don't know they're there until a search fails to find an obvious word, spell-check flags something normal, or a line wraps in a strange place. If that sounds familiar, read our explainer on invisible characters in AI text — or jump straight to the ChatGPT Watermark Remover to strip the hidden set in one click.
Where a broken ChatGPT paste hurts most
Email. Markdown asterisks render as literal symbols to your reader, and a non-breaking space can hold two words hostage on one line. Clean before you send.
Reports and documents. Word's styles and find-and-replace assume plain characters; smart quotes clash with the rest of the file, and a stray zero-width space makes spell-check flag ordinary words.
CMS drafts and web pages. Hidden characters land in slugs and metadata, surfacing as ugly %E2%80%8B sequences in URLs and breaking on-site search.
Translation and code. Translation tools segment on clean text, and compilers read bytes, not intentions — a byte order mark before a brace is not valid JSON, and a zero-width joiner inside a string literal is a bug you cannot see.
Remove ChatGPT formatting the manual way
You can clean a paste by hand with find and replace:
- Remove the Markdown markers. Replace
**with nothing, then*, then backticks, then the leading#characters. Each symbol is a separate search. - Fix the quotes. Replace the curly opening quote, the curly closing quote and the curly apostrophe with straight versions. That's several more searches, and it's easy to miss one variant.
- Fix the dashes and ellipses. Replace the em dash with your preferred dash style, and the ellipsis character with three periods.
- Hunt the invisible characters. This is the step that stops most people: you cannot search for a character you can't type. Word's wildcard search can find some of them, but you need the Unicode code points to do it.
For a two-line paste this takes a minute. For a full report it's ten minutes of find-and-replace whack-a-mole, and something still slips through.
If the junk came from a document or a web page rather than a chat — fonts, colors, styles, HTML tags — the Remove Formatting tool strips that whole layer in one click, no find-and-replace needed.
The one-click way
The AI Text Cleaner removes all of it in one pass. Paste the text you copied from ChatGPT, Claude, Gemini or any other AI chatbot, and it:
- strips Markdown symbols like
**,##, backticks and list markers - converts smart quotes and apostrophes back to plain straight quotes
- converts em dashes, en dashes and ellipsis characters to plain equivalents
- removes zero-width spaces, byte order marks and directional marks
- turns non-breaking spaces back into normal spaces
- repairs the broken spacing that invisible characters leave behind
Everything runs in your browser. Your text is never uploaded, there is nothing to install, and no account is required.
Before and after: one paste, one click
What you copy out of the chat window:
## Project update
**Reminder:** the report — "final version" — is due Friday…
After one pass through the cleaner:
Project update
Reminder: the report - "final version" - is due Friday...
If your paste also contained invisible characters, the two blocks above would look identical, which is exactly why they are so hard to remove by hand. The cleaner removes them too.
A quick pre-paste checklist for ChatGPT text
- Paste into the cleaner first, then into Word, Docs or your email draft
- If you paste directly, "paste as plain text" helps with fonts and colors, but remember it does not remove literal Markdown characters
- Scan for leftover
#,*or backticks before you hit send - If search or spell-check behaves strangely after pasting, invisible characters are the likely cause; run the text through the cleaner once more
Clean text should be the default, not the reward for fifteen minutes of find and replace. Paste, click, done. More guides like this one live on the AI text cleanup guides page.
Common questions about removing ChatGPT formatting
Why does copied ChatGPT text carry these symbols at all?
ChatGPT writes answers in Markdown, and the copy button often grabs the raw source — markers included. On top of that, the models learned from web pages and PDFs full of typographic Unicode, so curly quotes, em dashes, and the occasional invisible character come along for the ride. None of it is visible in the chat window; all of it lands in your document. One cleanup pass removes exactly that layer.
Why doesn't "paste without formatting" fix it?
Because the Markdown markers are characters inside the text, not formatting applied to it. "Paste as plain text" and Word's "keep text only" strip fonts and colors — they leave ** and ## exactly where they were. The same goes for invisible characters: they're characters, so they paste straight through. Removing them takes a text-level pass, which is what the cleaner does.
Does cleaning change what my text says?
No. The cleaner removes or normalizes characters — Markdown markers, smart quotes, em dashes, hidden characters — and never rewrites a sentence. Your words stay in place, in order, reading exactly the same. Fenced code blocks pass through byte-identical, and every rule is a switch you can turn off. The before/after character counts show precisely what was removed.
What about the hidden watermark characters?
The cleaner strips zero-width spaces, non-breaking spaces, byte order marks, and directional marks as part of every pass. If you want to see them before you clean, the Invisible Character Viewer reveals each one with its code point and name. And if hidden characters are the whole problem, the ChatGPT Watermark Remover is the focused tool for exactly that set.
Is it free? Is there a word limit?
Free, and no limit. The cleaner is a small set of character rules that run entirely in your browser — there is no server cost to meter and nothing to gate behind an account. Paste, clean, copy, as often as you like; the text never leaves your device, so it is safe for drafts and internal documents.