Paste a ChatGPT answer and get plain text back: bold, headings, code fences, tables, citation markers like 【12†source】 and [1], hidden characters and reasoning blocks are all removed in one pass.
Paste the answer into the editor above. The markdown ChatGPT writes in (**bold**, # headings, backticks, tables, > quotes) is reduced to plain text, the citation markers it leaves behind (【12†source】, [1], oaicite tags) are removed with the spacing tidied, hidden characters are stripped, and a DeepSeek-style <think> block is dropped. Click Copy Result and paste anywhere. Nothing leaves your browser.
Use the copy button under the reply or select the text. Both work; the tool handles the markdown either way.
Formatting, citation markers, hidden characters and any <think> block are removed as you paste, and the status line counts each kind.
Plain text is the default. Open the preferences to keep • bullets, straighten quotes or replace em dashes.
Into Upwork, a CRM, a form, a plain-text email or anywhere else that shows markdown symbols literally.
ChatGPT has no plain-text copy option. The copy button under an answer puts both rich text and markdown on the clipboard, and selecting the text with the mouse copies the markdown, so either way you arrive with bold markers, headings or a gray background. Paste the answer here instead: the editor reduces it to plain text and Copy Result gives you a clean version to paste anywhere.
Citations. When ChatGPT searches the web or reads a file it marks the sentence it is citing, and that marker survives the copy as bracketed text — 【12†source】, [oaicite:0], a leaked citeturn token, or a plain [1]. Perplexity uses [1] and [2] for the same job. This page removes all of them and tidies the space left behind, so "the figure rose [1]." becomes "the figure rose."
Yes, by giving you plain text. The gray box is styling carried in the rich-text copy, and plain text carries no styling at all, so there is nothing for Word to show. If you want real bold and headings in Word rather than plain text, use the ChatGPT to Word converter, which strips the background but keeps the structure.
The heading text stays on its own line; only the # markers go. Lists keep their line breaks and a hyphen bullet, or a • bullet if you turn on Keep bullets in the preferences. Numbered lists are unchanged. What you lose is the styling, not the structure.
The cleaner's job is hidden characters and spacing: it removes zero-width characters and the narrow spaces ChatGPT hides in text while leaving your bold and bullets optional. This page goes further and takes everything down to plain text, including italics, code fences, tables, citation markers and reasoning blocks. Use the cleaner when you want to keep the look; use this when you want the words only.
No. Everything runs in your browser with client-side JavaScript. The text is never sent to a server, stored or logged, which matters when the answer you are cleaning contains client names, contract terms or anything else you would rather not paste into a stranger's website.
An answer in the ChatGPT window looks like a tidy document: bold labels, headings, bullet lists, the occasional table. Underneath, it is three separate layers, and each one causes a different problem when you copy it somewhere else.
**bold**, headings as ## Heading, tables as rows of pipes. Anywhere that does not render markdown shows the symbols.There is no “copy as plain text” in ChatGPT, so you cannot opt out of any layer at the source. This page strips all three after the fact and reports what it found.
The markers are the part most tools miss, and the part that makes a pasted answer look obviously generated. The table lists the forms we see in copied text and what this page does with them.
| Marker | Where it comes from | Here |
|---|---|---|
【12†source】 | ChatGPT with browsing or file search; the bracketed source marker after a sentence | Removed |
:contentReference[oaicite:0]{index=0} | ChatGPT answers copied from a Project or from the API | Removed |
citeturn0search0 (with invisible wrappers) | ChatGPT's internal citation token when it leaks into copied text | Removed |
[1] [2, 3] [4–6] | Perplexity and ChatGPT search answers; attached to the sentence they cite | Removed |
[^1] and [^1]: … | Markdown footnotes in longer answers | Removed |
<think> … </think> | DeepSeek R1 and other reasoning models when the thinking is copied with the answer | Removed |
[1] Title — URL (a sources list) | The reference list some tools append at the end | Kept |
A numbered sources list at the end keeps its numbers on purpose, so you can still read it or delete it as a block. Inline markers next to the text they cite are what get removed.
Removal also tidies the gap. “Revenue grew 12% [1].” becomes “Revenue grew 12%.” rather than leaving a space before the full stop, and a pair like [1][2] is treated as two citations in the count. If your text contains an array index such as arr[0] outside a code span, switch Remove citations off in the preferences for that paste.
Models that show their reasoning — DeepSeek R1 is the common one — sometimes copy the whole thinking section with the answer, wrapped in <think>tags. It can run to several paragraphs of the model talking to itself. This page removes the block and the blank lines it leaves, and the status line notes “1 reasoning block removed” so you know it was there. If you want to keep the reasoning, paste it separately.
Plain text does not mean flattened text. Paragraph breaks, line breaks and blank lines are preserved. Numbered lists keep their numbers. Bulleted lists keep a hyphen (or a • if you turn on Keep bullets). Links become text (url), so both the wording and the address survive. The content of code blocks is kept exactly, with the fence lines removed. Em dashes and curly quotes are left as they are unless you change the preferences, where you can replace the dashes with a comma or hyphen and straighten the quotes in the same paste — the em dash remover explains why you might want to.
Upwork, Fiverr and job applications. Message boxes and cover-letter fields are plain text. A proposal with asterisks and hashes reads as unedited AI output, and clients have learned to skip those. Strip here, then follow the Upwork proposal workflow to make the content itself land.
CRMs, ticketing systems and internal tools. Salesforce notes, Zendesk macros, Jira comments in plain mode, intranet forms: all of them show markdown raw and most strip rich text. Plain text is the one format that arrives exactly as sent.
Plain-text email and SMS. Cold-email tools that send in plain text, and anything going out as a text message, need the words only. The cold email guide covers the writing side.
Spreadsheets. A table from ChatGPT comes out of this page as tab-separated rows, which paste into Excel or Google Sheets as real columns — often the fastest way to get a generated table into a sheet. For a typed .xlsx with number and currency formats, use the ChatGPT table to Excel converter.
Stripping is right when the destination cannot show formatting. When it can, convert instead. For Word and Google Docs, the ChatGPT to Word and Google Docs converters turn the markdown into real headings and lists and drop the gray background. For LinkedIn, Facebook and other plain-text social boxes, the AI text formatter keeps bold as Unicode bold and bullets as real bullets. For a markdown file, the ChatGPT to Markdown page gives you clean, standard markdown rather than removing it.
The clean-up runs in your browser with client-side JavaScript. Your text is never sent to a server, stored or logged. No account, no limits, no watermark.