Why ChatGPT answers are full of asterisks
You copy an answer from ChatGPT, paste it into an email or a web form, and get **Key point:** and ### Next steps instead of bold text and a heading. Claude and Gemini answers do the same thing.
Nothing is broken. Those symbols are formatting instructions written in Markdown, a plain-text format. The chat window reads the instructions and shows you bold text. The place you pasted into either doesn't read Markdown or never received the formatted version.
Remove the asterisks from AI textPaste your answer here
The quick fix depends on where the text is going
Decide first whether the destination can show formatting at all.
It can show formatting (Word, Google Docs, Gmail, Outlook). You want the symbols turned into real bold, headings and lists, not deleted.
- In Google Docs, check Enable Markdown under Tools, Preferences, then right-click and choose Paste from Markdown. Google's help page covers headings, bold, italics, strikethrough and links.
- For Word, email or anything else, paste the answer into MDMelt, pick the destination, and copy the result. The symbols become formatting before they reach the clipboard.
- The guide on keeping formatting in Word and Google Docs covers the built-in options in more detail.
It can't show formatting (text messages, most web forms, plain-text notes, app store reviews). Here the symbols need to go. Deleting every asterisk by hand works for a short reply and gets tedious fast. MDMelt's Plain text option removes them all, turns list markers into bullets, and rewrites each table row as a line of text, since a form field can't hold a table.
It's a chat app (Slack, WhatsApp, Telegram, Discord). Each one has its own formatting marks, and they don't all match Markdown. WhatsApp uses one asterisk for bold, not two. See pasting AI text into Slack and WhatsApp.
You'd rather not get Markdown in the first place. Say so in the prompt: "Reply in plain sentences with no Markdown symbols." Anthropic's prompting guide suggests stating what you want instead of what you don't, for example "Your response should be composed of smoothly flowing prose paragraphs." It only lasts for that conversation, and answers that needed a list or a table lose those too.
What the symbols mean
| You see | It means |
|---|---|
**text** or __text__ | Bold |
*text* or _text_ | Italic |
***text*** | Bold and italic |
#, ##, ### at the start of a line | Heading, level 1, 2, 3 |
-, * or + at the start of a line | Bullet point |
1. at the start of a line | Numbered list item |
> at the start of a line | Quoted text |
`text` | Code or a literal value |
[text](https://example.com) | A link with text in front of it |
~~text~~ | Strikethrough |
--- on a line by itself | A horizontal divider |
| Lines of pipes with a row of hyphens under the first | A table |
A table looks like this before it is rendered:
| Plan | Price |
|-------|-------|
| Basic | 10 |
| Team | 25 |
Math is a separate layer. AI answers usually write formulas in LaTeX, so you may also see things like \frac{a}{b} or \(x^2\).
Why AI chat apps write in Markdown
A language model produces text, character by character. It can't hand the app a bold font. What it can do is write characters that mean "make this bold", and Markdown is a widely known set of such characters.
John Gruber released Markdown in 2004 as "an easy-to-read, easy-to-write plain text format", with the goal that a Markdown document stays readable even when nobody renders it. CommonMark later pinned down the exact parsing rules, and GitHub's variant (GFM) added tables and strikethrough on top.
The AI companies treat Markdown as the default on purpose. OpenAI's Model Spec, in its December 2025 version, says: "Unless otherwise specified, assistant outputs should be formatted in Markdown with LaTeX extensions." Anthropic's documentation for developers has a section on asking Claude to use less Markdown, which tells you it is the habit to begin with.
So the asterisks are not a glitch that will get patched. They are how the answer is written underneath.
Why the symbols survive the paste
When you copy something, the clipboard can hold several versions of it at once. MDN's documentation of the browser clipboard gives the example of rich text "stored as both text/html and text/plain". The app you paste into picks the version it understands.
- Word, Google Docs and most email editors take the HTML version, so formatting can come through.
- Notepad, text messages, most form fields and many chat inputs take the plain-text version.
For AI answers, the plain-text version is often the raw Markdown, and the copy button in particular tends to copy it. OpenAI doesn't document exactly what its copy button puts on the clipboard, and the result has differed between the web app, the desktop apps, and selecting the text by hand. If one method gives you symbols, the other may not.
A note on Chinese, Japanese and Korean text
Sometimes the asterisks stay even inside apps that do read Markdown. CommonMark decides whether ** can open or close bold by looking at the characters on either side. A closing ** that has punctuation just inside it and a letter just outside it doesn't count. The spec's own example is a**"foo"**, which stays as literal asterisks.
English nearly always has a space or punctuation after the closing marks, so this rarely shows up. Japanese, Chinese and Korean don't put spaces between words, so **「重要」**は or **18%**늘었다 stays unformatted in a strict CommonMark renderer. MDMelt reads Markdown with the markdown-it-cjk-friendly rules, which fix this case for East Asian text and leave other languages alone.
Menu names and features in this guide were checked on November 15, 2026.
Sources
- Markdown, John Gruber (Daring Fireball)
- CommonMark Spec 0.31.2, emphasis rules and example 380
- GitHub Flavored Markdown Spec (tables)
- OpenAI Model Spec, December 18, 2025 version
- Prompting best practices, control the format of responses (Anthropic)
- ClipboardItem (MDN Web Docs)
- Use Markdown in Google Docs, Slides & Drawings (Google Docs Help)
- markdown-cjk-friendly (GitHub)