What Is the Markdown Prompt Cleaner?

The Markdown Prompt Cleaner takes a prompt that is full of formatting and returns clean plain text. It removes the things that waste attention when you paste text into a chat model: Markdown headings and emphasis, fenced code blocks, HTML tags, emoji characters and web links. You decide what gets removed with five independent switches, so you can strip everything or keep one category and drop the rest.

The point is simple. Text copied from a document, a wiki page, an email or an old chat often carries syntax the model does not need. Cleaning it first means fewer wasted characters, a clearer instruction and a prompt you can paste anywhere without strange symbols showing up in the answer.

Why You Would Use It

If you work with prompts daily you probably keep recycling text: a brief from a teammate, a section of a spec, a support transcript, a style guide copied from a website. Each source brings its own junk. Markdown stars, angle brackets, tracking links and reaction emoji all take up room and can be read as instructions rather than content.

This tool is for prompt engineers tidying a library of prompts, marketers moving copy from a CMS into a model, developers turning a README section into an instruction, teachers preparing clean reading material, and anyone who has pasted a wall of formatted text and watched the model echo the formatting back. It is also handy for checking how long a prompt really is in characters, words and approximate tokens before you send it.

How This Tool Works

Paste the raw prompt into the editor at the top. The editor is a code style pane with Markdown syntax support, and it has Paste, Sample and Clear buttons so you can load a realistic example in one click. Under it, five toggles control the cleanup. All five start switched on.

When you press Clean prompt, the tool applies the local removals first and in a fixed order. Fenced code blocks delimited by three backticks or three tildes are deleted whole, including the code inside, and the resulting blank gaps are collapsed. Then HTML tags are stripped while the text between them is kept. Then anything starting with http, https or www is removed. Then emoji characters, variation selectors and skin tone modifiers are stripped.

Only after that does the Markdown step run, and only if you left it switched on. Your text is sent to the tool Markdown endpoint, converted with a CommonMark parser into HTML, reduced to text with the tags removed, decoded from HTML entities, and normalised: line endings become single newlines, runs of spaces collapse, and three or more blank lines become one blank line. The result appears in the output box with a live count of input characters, output characters, output words and an approximate token figure (output characters divided by four).

If the server step fails, an error toast appears, the status line clears and the previous output stays untouched, so you never copy a half cleaned prompt. If you selected no options at all, the tool warns you to select at least one before it does anything.

Key Features

• Five independent toggles that can be combined in one pass
• Accurate Markdown to text conversion through a CommonMark engine
• Code fence removal that deletes the whole block, not just the markers
• HTML stripping that keeps the readable text between tags
• URL removal for http, https and www links
• Emoji removal including variation selectors and skin tone modifiers
• Character, word and approximate token counts for input and output
• Sample, Paste, Clear and a Copy button that unlocks once there is output

How to Use the Tool

1. Paste your prompt into the Raw prompt editor, or click Sample to load an example that exercises every option, or Paste to read from your clipboard.
2. Review the five toggles under What to remove. Turn off anything you want to keep; for example leave Remove code blocks off if the code is part of the instruction.
3. Press Clean prompt. The status line shows Cleaning while the Markdown step runs, then Done.
4. Read the counts under the output, check the text, then press Copy clean prompt to put it on the clipboard.
5. Press Clear to empty both boxes when you start the next prompt.

Limits and Rules

The Markdown conversion step accepts up to 200,000 characters. Longer input is rejected, so for very large documents clean the text in sections. The four local removals have no practical size limit because they run in your browser.

Removing code blocks is destructive: the entire fenced block disappears, including the code. Disable that option if you want to keep code as text. Removing URLs deletes the whole link, not only the protocol, so a reference that matters should be kept by turning Remove URLs off. Removing Markdown flattens headings, bold, italic, links, lists and inline code into plain text, which means the structure is gone from the output.

Only the Markdown step sends text to the server, over HTTPS; the other four run locally. The endpoint does not store what you send, and this tool route carries no request-tracking middleware either, so no route, status or timing record is written for your call. Do not paste confidential content into any online tool you are not permitted to use.

When to Use It and When Not to

Use it before sending a copied brief, a documentation excerpt, a transcript or a support thread into a chat model, when you want a shorter prompt with no stray symbols, or when you need a token estimate for a prompt you are about to reuse many times.

Do not use it when the formatting is the point. If you are asking a model to rewrite Markdown, review HTML or edit code, stripping those first removes the material you wanted to work on. It also does not summarise, rewrite or restructure text; it only removes what you told it to remove, and it cannot judge whether a link or a code block was important.

Side Effects to Check Before You Copy

Each switch is a blunt pattern, not a judgement about meaning. Most of the time that is exactly what you want, but a few cases catch people out, especially with prompts that were written carefully in the first place.

Remove HTML also removes prompt tags and placeholders

The HTML step deletes anything that sits between a less-than and a greater-than sign. That includes XML-style sections many people use to structure prompts, such as <instructions> and <context>, and template placeholders like <customer_name>. The text between paired tags survives, but the tags that told the model where each section starts and ends are gone. A sentence such as "if a < b and c > d" also loses everything between the two signs. Turn Remove HTML off when your prompt relies on angle-bracket tags.

Remove markdown flattens lists and tables

Bullets and numbers are dropped from lists, so "1. first" becomes "first". A Markdown table is broken into one cell per line, with no column labels next to the values. If order or a table layout matters to the instruction, rewrite that part as sentences before cleaning, or leave Remove markdown off.

Remove URLs runs before the Markdown step

A link such as [docs](https://example.com) first loses its address and then, if Remove markdown is on, is reduced to the word "docs". If Remove markdown is off, the leftover [docs]() stays in the output. A full stop or comma stuck to the end of a bare URL is removed together with it.

Remove emojis catches some symbols too

The emoji pattern matches pictographic characters, which include the copyright, registered and trademark signs and symbols such as a heart or a double exclamation mark. A keycap emoji like 1️⃣ can leave its digit plus a combining mark behind. Plain symbols such as a check mark, star or arrow are left alone.

An unclosed code fence removes the rest of the text

Code-block removal deletes everything from an opening fence to the next fence of the same kind. If the closing fence is missing, every line after the opening fence disappears. Compare the input and output character counts; a much shorter result than expected usually points to this.

Frequently Asked Questions