Every Markdown renderer answers the same question differently: what will this text look like after someone else renders it? This page records the rules the Markdown Preview on this site applies, so you can tell a renderer problem from a document problem. The statements below were checked against the tool component and against micromark 4.0.2 with micromark-extension-gfm 3.0.0 on 2026-11-03. Every HTML fragment quoted here is the renderer's output for that input.

CommonMark, GFM, and What This Preview Renders

CommonMark 0.31.2, published 2024-01-28, is the baseline: paragraphs, ATX and setext headings, block quotes, ordered and unordered lists, fenced and indented code, inline and reference links, images, emphasis, code spans, thematic breaks, backslash escapes and entity references.

GitHub Flavored Markdown 0.29-gfm, published 2019-04-06, is a strict superset of CommonMark. Its specification adds five extensions: tables (section 4.10), task list items (5.3), strikethrough (6.5), autolink literals (6.9) and disallowed raw HTML (6.11). GFM also states in section 1.1 that GitHub applies extra post-processing and sanitization after conversion.

The preview runs micromark with its GFM extension and turns on every extension that package ships: the five from the specification plus footnotes. In practice you get:

  • Tables with column alignment.
  • Task lists. The checkboxes are disabled, so you can read the state but not change it in the preview.
  • Strikethrough with one or two tildes; ~one~ and ~~two~~ both produce del elements.
  • Autolink literals: a bare https://example.com, a www.example.com host and an email address each become a link, the last one with mailto:.
  • Footnotes with a superscript reference and a backlink.

There is no button to press: the preview renders 300 ms after the last edit, so a fast typist sees one render per pause rather than one per key.

What It Does Not Render

The renderer implements CommonMark and the GFM extensions. Anything outside those two specifications has no meaning here, and the source text stays visible:

  • Math: $x^2$ and a $$ block remain text; no formula is typeset.
  • Mermaid and other diagram fences: a mermaid block becomes an ordinary code block. There is no diagram and no syntax color.
  • Emoji shortcodes, user mentions and issue references, such as :tada:, @octocat and #123, stay as typed.
  • GitHub alerts: > [!NOTE] stays a block quote whose first line is the literal marker. The block quote itself still renders.
  • Heading anchors: no heading receives an id. The only identifier the renderer emits is footnote-label on the visually hidden footnotes heading.
  • Front matter, definition lists, includes, and table-of-contents markers: not stripped, not parsed, and not replaced.

A document written for one platform often contains one of these forms; the preview leaves it as text, so the difference is visible before you publish.

Raw HTML Is Escaped

The renderer is created with allowDangerousHtml: false, so every HTML tag in your Markdown is escaped and displayed as text. This input:

<details>
<summary>More</summary>

Hidden line.

</details>

produces this output:

&lt;details&gt;
&lt;summary&gt;More&lt;/summary&gt;
<p>Hidden line.</p>
&lt;/details&gt;

The paragraph inside is still parsed, because a blank line starts a new block after the escaped tag lines. The containers are not.

The reason is the origin. The preview runs on zerotool.dev, the same origin as tools that keep their inputs in browser storage. If raw HTML were rendered, an img tag with an onerror attribute in pasted text would run as script on this site, with access to whatever those other tools stored. Escaping removes that path completely: no tag from your text becomes a live element.

GitHub takes a narrower route. GFM section 6.11 escapes the opening angle bracket of nine tag names (title, textarea, style, xmp, iframe, noembed, noframes, script, plaintext), and GitHub then sanitizes the result, as section 1.1 says. This preview escapes every tag, which is stricter than both steps.

Links and images follow the same principle. A link keeps its address only if it is relative or uses http, https, mailto, irc, ircs or xmpp. An image keeps its address only if it is relative or uses http or https. Everything else is emptied while the link text stays:

[docs](javascript:alert(1)) and ![x](data:image/png;base64,AAAA)

renders as:

<p><a href="">docs</a> and <img src="" alt="x" /></p>

So write links, images and emphasis with Markdown syntax. For a line break inside a paragraph, use a backslash at the end of the line or two trailing spaces; both produce a br element, and neither depends on raw HTML. If a document truly needs details, kbd or sub, check it on the platform that will publish it.

Copy HTML copies this escaped output, without the preview's syntax colors and without the target and rel attributes the preview adds. Raw HTML in the copy stays escaped, so a page that pastes it receives no element from your text beyond what Markdown itself produces.

When Bold Fails Next to CJK Punctuation

This is the case that surprises Chinese, Japanese and Korean writers most often. Type this:

**注意:**这里

and the preview shows the asterisks:

<p>**注意:**这里</p>

Nothing is broken. CommonMark decides emphasis with delimiter runs, and section 6.2 defines a right-flanking run as one that is not preceded by whitespace and either is not preceded by punctuation, or is preceded by punctuation and followed by whitespace or punctuation. Here the closing ** is preceded by :, a Unicode punctuation character, and followed by 这, a letter. Neither branch holds, so the run cannot close, and the opening run stays literal too.

Three edits fix it, and they all keep the meaning:

  • Add a space after the closing marks: **注意:** 这里 gives <strong>注意:</strong> 这里.
  • Add an ideographic space (U+3000): **注意:** 这里. U+3000 is in Unicode category Zs, which CommonMark counts as whitespace, so the run closes and the wider space is kept.
  • Move the punctuation outside: **注意**:这里 gives <strong>注意</strong>:这里.

The rule is not about CJK characters; it is about punctuation touching a letter. English has the same failure: **Note:**this stays literal, while **Note:** this is bold. CJK text simply hits the pattern more often, because a full-width colon or bracket normally sits directly against the next character with no space.

The same cause explains **「重要」**です and **10%**の割引: the closing run follows a closing bracket or a percent sign and precedes a letter. Add a space, or move the punctuation outside the marks.

This site's Markdown to Word converter enables the CJK-friendly micromark extensions, so it accepts these runs without a space. The preview deliberately does not: its job is to show what a CommonMark or GFM renderer will do with the same text. The Markdown cheat sheet lists the same cases with their rendered output.

Code, Tables, Lists, Quotes, and Front Matter

A fenced block takes a language tag after the opening fence, and the tag selects the colors:

const greet = name => `Hello, ${name}!`;
console.log(greet("ZeroTool"));

Twelve languages are registered, under these names: JavaScript (js), TypeScript (ts), Python (py), Go, PHP, Ruby (rb), Java, Bash (sh, shell), JSON, XML (html), CSS and Markdown (md). Any other tag produces an ordinary code block: the text is escaped and shown in monospace with no colors. Indented code (four spaces) also works and never has a language.

Tables need a header row and a delimiter row. Colons in the delimiter row set alignment, and a pipe inside a cell is written \|:

| Setting | Default |
|:--------|--------:|
| retries | 3 |

The output carries the alignment on every cell, as align="left" and align="right" in this example. A leading and a trailing pipe are optional; the delimiter row decides how many columns the table has.

Nested lists follow the content column of the parent item. A bullet marker plus one space is two columns, so two spaces before the nested marker are enough:

- Parent
  - Child

An ordered list keeps the number it starts with: a list that begins with 3. gets start="3" in the output. Block quotes nest with a second >; a blank quoted line starts a second paragraph inside the same quote.

Front matter is not part of CommonMark or GFM, and the preview does not strip it. A file that begins with a --- line gets a thematic break, and the closing --- acts as a setext heading underline for the line above it:

---
title: "Release notes"
---

Body

renders as:

<hr />
<h2>title: &quot;Release notes&quot;</h2>
<p>Body</p>

The practical fix is to delete the front matter before pasting, or to put the whole file inside a fenced block when you want to inspect it. Setext headings (Title over ====) and reference-style links do work, so a document that uses them renders as expected.

Footnotes

Footnotes are enabled even though the GFM specification does not define them; GitHub added them to Markdown fields on 2021-09-30, and micromark's GFM bundle includes an implementation. The syntax is a reference in the text and a definition elsewhere:

A note.[^1]

[^1]: The note text.

The first line becomes a superscript link with the identifier user-content-fn-1, and the definition becomes a section with class footnotes, an ordered list and a backlink. The generated heading is the English word "Footnotes", marked sr-only, on every language version of the page; the tool has no localized label for it. Because the reference is a fragment link, clicking it scrolls inside the preview instead of opening a new tab.

Definitions can appear anywhere in the document, including before the paragraph that references them, and one definition can be referenced more than once. The numbering follows the order of first reference, not the order of definitions.

Privacy: Rendering in This Page

The Markdown is parsed and rendered by JavaScript in the tab. It is not uploaded, and the tool does not save it: the component never calls the site's storage helper, so the text is not written to local storage, a cookie or the address bar by the tool. Clear empties the editor, the preview and the count, and cancels a pending update; Ctrl/⌘+L does the same when focus is inside the tool.

Images are the exception. ![alt](https://…) creates an image element that your browser fetches from that address, so the image host sees your IP address and browser details. A relative path such as images/a.png is requested from zerotool.dev, where the file does not exist; a file: address is emptied. If the text must not cause any request, keep images out of it or use a placeholder.

Copy HTML records one analytics event after the clipboard write succeeds, with the tool name and the action only. A failed or superseded copy records nothing, and neither the Markdown nor the rendered HTML is part of the event. The tool page loads the site's analytics and ad scripts like other tool pages; the Markdown is not an input to either.

Where GitHub, GitLab, and Notion Differ

The preview shows the CommonMark and GFM layer. Other products put their own layer on top, and that layer is where most surprises come from.

GitHub converts GFM and then post-processes and sanitizes the result (GFM section 1.1). It generates heading anchors, so links like #installation work there and not here. In issues and pull requests, task list boxes are interactive, while this preview keeps them disabled. GitHub-only syntax such as alerts, math, Mermaid diagrams, emoji shortcodes, mentions and issue references has no CommonMark meaning, so a document that uses it looks plain here.

GitLab's own documentation describes GitLab Flavored Markdown as CommonMark plus the GFM extensions plus extensions specific to GitLab, and lists front matter, alerts, description lists, math, diagrams and GitLab references among the additions. It also generates a heading anchor for every heading. A file whose front matter this preview turns into a horizontal rule and a heading is metadata in GitLab.

Notion is a block editor. Its help describes Markdown as shortcuts used while typing, and a page is made of blocks rather than Markdown text. A direct comparison therefore holds only for text that survives as block content. Toggles, databases and synced blocks have no Markdown form, so there is nothing for a Markdown renderer to show.

The useful rule: this preview answers "what does the CommonMark and GFM layer produce", and platform layers are additive. When a document must look right on one platform, render it there before publishing.

The word count splits the trimmed source on whitespace, so a Chinese or Japanese paragraph counts as one word. The character count is the length of the string in UTF-16 code units: it includes spaces and line breaks, and a character outside the basic plane counts as two. The preview has no file picker and no export; the editor is a plain text area.

Copy HTML gives the renderer's HTML without the preview's color spans and without the target and rel attributes, which the preview adds to open non-fragment links in a new tab. Raw HTML stays escaped in the copy. Images keep their original addresses, so a page that pastes the copy will also request them from those hosts.

Related tools: the Markdown Table Generator builds tables with escaped pipes and aligned columns for Chinese, Japanese and Korean text; the Markdown to Word converter exports a document with heading styles, numbering, tables and footnotes, and recognises bold next to CJK punctuation; the Markdown cheat sheet lists the syntax with the HTML it produces. The Markdown Preview itself is the place to check a document before you paste it anywhere else.