Back to list
Text5 min read

Remove ChatGPT Markdown Symbols (**, #)

Remove the **, # and table symbols from copied ChatGPT answers before pasting into Word, chat or email, step by step.

By webgyver

Why Do Asterisks and Hash Marks Follow Along: Markdown in Five Minutes

ChatGPT and many other AI chatbots write answers in Markdown, a lightweight set of notation rules. Markdown mixes symbols into plain text to mark formatting. It was introduced by John Gruber, and the CommonMark specification later spelled out the syntax in detail. In these rules a # at the start of a line means a heading, with one to six allowed and the count giving the heading level. Text wrapped in two asterisks is bold, and text wrapped in one asterisk or underscore is italic.

In the chat window these symbols are interpreted and appear as bold text and large headings. If you paste the copied content into an input box that does not interpret Markdown, the symbols show up as literal characters. The result can depend on how you copied and where you paste, so the same answer may arrive as formatting in one place and as symbols in another.

The symbols you meet most often are these. A # at the start of a line is a heading, ** is bold, a hyphen or a number with a period at the start of a line is a list, and [text](address) with brackets and parentheses is a link. A table drawn with pipes (|) and hyphens is an extension called GFM (GitHub Flavored Markdown), and so are ~~strikethrough~~ and checkbox lists. That is why an answer that comes with a table is the hardest to clean up.

For example, if an answer contains a line like ## Summary, it is a level-2 heading that begins with two hash marks, and a part written as **important** is meant to be shown in bold. A screen that interprets Markdown shows only the formatting with the symbols gone, while a place that does not interpret it shows every symbol as text. So the problem is not that the answer is wrong. It comes from a side that understands the notation meeting a side that does not, and a quick cleanup suited to the destination solves it.

Handling It by Destination: The Method Depends on Where You Paste

Start with pasting into Microsoft Word. First separate two problems. One is that formatting such as fonts and colors comes along when you copy from the web. The other is that asterisks and hash marks arrive as characters. Word's paste options are Keep Source Formatting (K), Merge Formatting (M) and Keep Text Only (T), and Keep Text Only discards formatting and nontext elements such as pictures and tables. On Windows the same action is available as Ctrl+Shift+V. But this removes formatting, not the ** characters, so symbols that already arrived as characters stay.

Leftover symbols can be removed with Word's Find and Replace (Ctrl+H). Enter ** in Find what and leave Replace with empty, with the Use wildcards option turned off. According to Microsoft's documentation, when wildcards are on an asterisk means any string of characters, and to search for the asterisk itself you type a backslash before it, as in \*. The same documentation advises using Replace one at a time rather than Replace All so you can confirm each change.

Hangul (Hancom Office) presumably has Find and Replace and a paste-without-formatting feature as well, but menu names and locations differ by version, so this article does not claim specific steps. The principle is the same in any program: strip formatting with a paste-without-formatting command, then handle leftover symbols with Find and Replace, or clean the symbols before pasting.

The input box of a messenger such as KakaoTalk is generally plain text that does not interpret formatting, so symbols are often sent exactly as they appear (this may vary by app version, so check for yourself). For the recipient, symbols such as ## get in the way of reading, so it is better to remove them before sending. In a long answer, numbers and hyphen lists still read well, so removing only heading marks and asterisks is usually enough.

With email, the result depends on whether the compose window is rich text or plain text. In rich text mode, pasted formatting can follow, so try pasting without formatting first. The symbols still remain, so they need separate cleanup. Conversely, on a site such as GitHub that renders GFM, pasting the symbols as they are is the most accurate. Before you delete anything, check whether the destination understands Markdown.

Manual Cleanup or a Tool: Which Is Better When

The advantage of manual cleanup is that you need nothing extra and you control the result by eye. A few asterisks and two or three heading marks are done in a handful of Find and Replace passes. But the longer the answer, the more kinds of symbols it contains. Heading # marks vary from one to six, which makes them hard to find in one pass. A hyphen at the start of a list item must be told apart from a hyphen inside a sentence, and links require handling brackets and parentheses together.

Removing everything at once with a regular expression saves repetitive work but has exceptions. According to the CommonMark specification, an underscore inside a word does not create emphasis, so the underscores in a name like foo_bar_baz are content, not formatting. To avoid deleting those, a parser that reads the document according to the rules is safer. In short, a line or two is quicker to fix by hand, while a long answer that mixes tables or code, or a batch of repeated cleanups, suits a tool better.

TheWebGyver's Markdown stripper is based on the remark parser and shows how many elements of each kind it removed, such as headings, bold text, links, code blocks and tables. Processing happens inside the browser, and the site states that your input is not sent to an external server. Even with a tool, it pays to skim the result once.

Step by Step: Five Steps to a Clean Conversion

Step 1: copy the answer you want to clean up from ChatGPT, whether with the copy button or by dragging. Step 2: paste it into the tool's input box. If it is your first time, the load-example button lets you see how it behaves first. Step 3: in the original preview, check the highlighted Markdown that will be removed. This is the moment to look for any symbol you did not want to lose.

Step 4: check the converted result and the statistics of removed elements, then press the copy button. Step 5: paste it into its destination. In Word or email, pairing it with paste-without-formatting clears both symbols and formatting. After pasting, check three things: that headings have not merged with body paragraphs, that numbers and lists remain where you need them, and that link addresses remain if you need them.

Cautions: Common Worries and What Actually Happens

One worry is that deleting the symbols loses the formatting. In a plain text input box the formatting is not displayed anyway, and where formatting is needed you usually reapply it with that program's styles after cleaning. In Word, for example, you remove the heading marks and then apply a heading style.

Tables need attention. Converting to plain text loses the column separation, and Microsoft explains that Word's Keep Text Only paste also turns tables into a series of paragraphs. If the table is the point of the answer, move it into a Word table or a spreadsheet instead of flattening it to text.

Removing symbols only tidies formatting and has nothing to do with whether the content is accurate. AI answers can be wrong, so verify numbers, quotations and links yourself. Also, if copied text counts more characters than you can see, invisible characters such as zero-width characters may have been mixed in. Such characters can also appear through web page copying or editor conversion, so you cannot conclude that an AI service inserted them. The blog's article on AI text watermarks covers this separately.

If you do this often, it helps to fix a decision order. First check whether the destination understands Markdown, and if it does, leave the text as it is. If it does not, clean the symbols and then paste without formatting. If there is structure-dependent information such as tables or links, save the original separately before converting. Then, if the result is not what you wanted, you can start over from the beginning.

References

Try It Yourself

Experience what you learned in this article with Markdown Stripper.

Markdown Stripper