Remove Markdown From Text — Free Online Markdown Stripper

What markdown symbols look like in raw AI output

Ask ChatGPT or Claude to write almost anything longer than a sentence, and the response contains markdown. Headings arrive as lines starting with one to six hash symbols. Bullet points use a leading hyphen, asterisk, or plus sign. Bold text wraps in double asterisks or double underscores. Italic text uses single asterisks or single underscores. Code appears either inline between backticks or in fenced blocks delimited by triple backticks, sometimes with a language tag like ```python. Links follow the [label](url) syntax. Blockquotes start with a greater-than sign. Tables use pipe characters as column separators.

These symbols are intentional. The model's training data contains enormous quantities of markdown — GitHub READMEs, Stack Overflow answers, Reddit posts, documentation sites — so markdown is statistically natural output for any structured response. The model is not doing anything wrong. The problem is purely contextual: markdown is useful in a renderer and noise everywhere else.

Notion, Obsidian, and similar tools export their content with markdown preserved. A page exported from Notion as plain text still contains # headings and **bold** spans. Obsidian's clipboard output is raw markdown by default. If you take that export into a CMS text field, an email body, or a plain-text API parameter, the symbols land in the output literally.

The nine markdown patterns and how each gets stripped

Headings: a line starting with one or more # characters followed by a space. The regex /^#{1,6}\s+/gm matches and removes them, leaving the heading text on its own line with no indentation change.

Bold: **text** or __text__. The delimiters are removed and the content survives. The safe version of this regex checks for word or space boundaries on both sides — more on why that matters in the next section.

Italic: *text* or _text_. Same boundary logic as bold. A single asterisk or underscore without a matching pair passes through unchanged, which is the correct behavior for stray punctuation.

Inline code: a span wrapped in single backticks. The backticks are stripped and the code text remains. This is almost always the right call — the reader needs the value, not the formatting signal.

Fenced code blocks: three or more backticks, optionally followed by a language identifier, enclosing one or more lines, closed by another triple-backtick fence. The fences and the language tag are removed; the code content is kept. This differs from inline code only in that the fence spans multiple lines.

Links: [label](url) collapses to label. The URL is discarded. If you need the URL to survive, copy the link address separately before running the tool.

Images: ![alt](url) collapses to alt. For most prose contexts, the alt text is the useful part.

Blockquotes: a leading > and optional space on each line. Removing the marker flattens the quote into a regular paragraph.

Tables: pipe-delimited rows and the separator line of dashes and colons. The entire table structure is removed in plain mode, leaving only the text values. In keep-lists mode, tables are also fully stripped — there is no meaningful plain-text equivalent for a table.

Escape sequences: markdown uses a backslash to escape its own special characters — \* renders as a literal asterisk rather than starting italic. After stripping formatting, the backslash is no longer needed, so \* becomes *, \[ becomes [, and so on.

Plain text vs. keep-lists: which output mode to choose

The tool offers two output modes. Plain text strips everything including list markers, turning each list item into a standalone line of prose with no leading character. This is the right choice when the destination is a paragraph-based field — an email body, a CMS rich-text editor that manages its own list formatting, or a form input.

Keep-lists leaves the leading dash and the text intact for unordered items and preserves the number-plus-period for ordered lists. Everything else — headings, bold, italic, code fences, links, blockquotes, tables — is still removed. Use this when the destination is a plain-text context that conventionally uses dashes for lists, like a Slack message, a README section in another format, or a plain-text export pipeline that processes list structure downstream.

Neither mode re-renders markdown as HTML. If you want markdown converted to HTML rather than stripped, that is a different operation that a browser-side markdown library handles. This tool is only for removing formatting symbols, not translating them.

Inline code vs. code blocks: why the handling differs

Inline code marks a short span of text — a variable name, a command, a value — inside a sentence. The backtick wrapper is a formatting signal. Stripping the backticks and keeping the text is almost always correct: `config.timeout` becomes config.timeout, which is readable and accurate.

Fenced code blocks typically contain multiple lines of actual code. The fence and the language tag are metadata; the content is the substance. Stripping the fences and keeping the code preserves what matters. The result lands as a contiguous block of lines in the output without the opening and closing fence markers.

One edge case worth knowing: nested backticks inside a fenced block are treated as literal characters, not as nested inline code spans. A block containing a shell command like echo `hostname` will preserve the inner backticks if they fall inside the fence. Once outside a fence, a bare backtick without a matching pair passes through unchanged.

Why naive regex breaks URLs and variable names

The two most common mistakes in markdown strippers involve underscores and asterisks in contexts where they are not markdown at all.

Consider a URL like https://example.com/api/get_user_profile. A naive /_([^_]+)_/g pattern targeting italic underscores will match _user_ inside that URL and remove the surrounding underscores, producing https://example.com/api/getprofile — a broken URL. The safe approach requires the pattern to check that the underscore is preceded by a space or the start of the line, not a slash or word character.

The same problem applies to Python and JavaScript variable names. A naive bold stripper applied to the string my_variable__name__here would match __name__ as bold and strip the underscores, yielding my_variablenamehere. Correct behavior: __name__ inside a word should pass through unchanged. The pattern needs a word-boundary or whitespace check on both the opening and closing delimiter.

A third case: asterisks in math. The expression 3 * 4 * 5 = 60 contains two asterisks that look like italic delimiters to a pattern that only counts pairs. A robust stripper checks that the closing asterisk is preceded by a non-space character and that the content between them does not contain a newline, which is how proper markdown parsers distinguish a multiplication expression from an italic span.

This tool applies boundary checks to all bold and italic patterns. It does not touch underscores that are surrounded by word characters on both sides, and it does not strip asterisks that appear in numeric expressions without whitespace-bounded pairs.

Common errors when stripping markdown manually

Deleting the asterisk used as a multiplication sign. In text like "Output: 3 * 4 = 12", a manual find-and-replace for * removes the operator. Always scan for math before running a global asterisk replacement.

Stripping underscores from variable names or file paths. A document that mentions __init__.py or snake_case_variable will be mangled by a global underscore strip. The correct fix is to remove only underscores that form balanced italic pairs around non-word-character boundaries.

Leaving pipe characters from tables. A table with three columns and five rows contributes about 20 pipe characters to the text. If the table stripper only removes the separator row (the |---|---| line) and not the data rows, the pipes end up in the output as literal characters scattered across the text.

Forgetting escape sequences. After stripping formatting, a literal \* should become *. If escape sequences are not processed, the output contains backslashes that were invisible in the original rendered markdown.

Double-processing nested structure. Running a bold stripper and an italic stripper in the wrong order on ***bold italic*** can leave stray asterisks. The correct order is: strip the combined bold-italic pattern first, then process remaining bold pairs, then remaining italic pairs.

Where markdown residue comes from: ChatGPT, Claude, Notion, Obsidian

ChatGPT renders markdown in its web and app interfaces, so users rarely see the raw symbols there. But copying from the ChatGPT web interface sometimes carries markdown into the clipboard — especially when copying a full response rather than running it through the chat interface's copy button. The API always returns raw markdown, so any application that calls the API and displays the response without running it through a renderer will show the symbols literally.

Claude by default uses markdown heavily in structured responses — numbered steps, bold key terms, code fences for any code snippet. The web interface renders it, but paste the response into a plain text area and the symbols are all there. Claude's system prompt can suppress some formatting, but it cannot be fully disabled through prompting alone for long responses.

Notion exports pages as markdown by default in its "Export as Markdown & CSV" option. The exported .md file contains full markdown, including heading levels, bold, italic, code blocks, and linked text. If you export to plain text instead, Notion still preserves some symbols for elements that have no plain-text equivalent. The cleaner handles both.

Obsidian stores all notes as markdown files and its clipboard behavior follows the same convention. Copying a section from Obsidian to another application pastes raw markdown. This is by design — the assumption is that the receiving application will render it. When it does not, this tool handles the cleanup.

For invisible characters that come from these same sources — zero-width spaces, soft hyphens, directional marks — the invisible character remover is the complementary tool. Running both in sequence cleans the output completely.

Privacy: your text stays in your browser

The stripper runs entirely in client-side JavaScript. No text is sent to a server at any point. There is no logging, no session storage of your content, and no analytics that capture text values. When you close or refresh the page, the input and output fields are cleared.

This is not a marketing claim — it is a structural fact. There is no server endpoint that receives text input from this page. You can verify this by opening your browser's network tab while using the tool: no outbound requests are made when you click Remove Markdown.

If you are cleaning sensitive content — internal documents, client data, proprietary code — the tool is safe to use without concern about data leaving your device. The same is true for the other tools on this site: the zero-width space remover, the extra spaces remover, and the special character remover all operate the same way.

When to use this tool vs. a markdown renderer

A markdown renderer — whether a library like marked.js, a CMS preview pane, or a chat interface — converts markdown syntax into HTML or formatted display output. The asterisks become bold text, the hyphens become visible bullet points, the hash symbols become headings with larger font sizes. The symbols disappear because they are interpreted, not removed.

Use a renderer when your destination can display HTML or when you want the formatting to survive in a different form. Use this stripper when your destination is plain text that cannot render formatting — a CSV field, a plain-text email, a command-line argument, a legacy API parameter, a form that counts characters and needs only prose.

A common workflow for AI-generated content: run the output through this tool to get clean prose, then paste the prose into your editor and apply its native formatting. You get the structure you want without inheriting the model's formatting choices.

For removing em dashes that AI models also insert frequently, the em dash remover on this site handles those. For stripping line breaks from PDF-exported text that has hard-wrapped lines, the line break remover is the right tool.