HTML to Markdown

Convert HTML to Markdown online — three ways in

This HTML to Markdown converter takes whatever you have and returns clean, GitHub-flavored Markdown, live as you paste:

• HTML source— paste markup (a fragment or a whole page from "view source") or drop a .html file.
• Paste rich text — copy formatted content from any web page, Google Docs, Word, Outlook, Gmail, or Notion and paste it; the formatting becomes Markdown.
• From a URL — convert a live web page to Markdown, main content only, with links made absolute.

Before converting, the HTML is cleaned: scripts, styles, SVG icons, forms, and hidden elements are dropped; lazy-loaded images get their real source; highlighted code blocks keep their language for the fence; empty permalink anchors next to headings disappear. Then Turndown with the GFM plugin writes the Markdown — tables, strikethrough, and task lists included. The preview tab renders the result so you can check it before you copy.

Rich text to Markdown: Google Docs, Word, and web pages

When you copy from a document or a web page, your clipboard holds HTML, not just text. The rich-text box reads that HTML, which is why formatting survives. Each source has its quirks, and the converter knows them:

• Google Docs to Markdown — Docs wraps everything in a <b id="docs-internal-guid…"> and expresses bold and italic as inline styles on <span>s, which naive converters turn into one big bold blob or lose entirely. Here they become **bold** and _italic_, strikethrough becomes ~~text~~, Google redirect links are unwrapped to the real URL, monospace text becomes `code`, runs of monospace paragraphs become a fenced code block, and tables get a proper header row.
• Microsoft Word— Word's clipboard HTML fakes lists with indented paragraphs and mso-list styles. Those are rebuilt into real, nested Markdown lists (numbered or bulleted), and the office-specific junk (<o:p>, conditional comments) is dropped.
• Web pages, email, Notion— select any part of a page, copy, paste. The result is the same Markdown you'd get from its HTML source, without having to open dev tools.
• Plain text to Markdown — pasting text with no formatting still works: paragraphs become Markdown paragraphs and single line breaks are kept.

For a whole Google Doc, Docs' own File → Download → Markdown (.md) is also handy; the paste route is quicker for a section, keeps working in docs you can only view, and lets you choose the Markdown style.

Web page to Markdown for LLMs and notes

Markdown is the format language models read best: it keeps the structure of a page — headings, lists, tables, code — at a fraction of the tokens HTML needs. Enter a URL and the page is fetched, reduced to its main content (the <article> or <main> element, or the densest block of text when a page has neither), and converted. With "Add title + source link" on, the Markdown starts with the page title and its URL, ready to paste into ChatGPT, Claude, a notes app, or a RAG pipeline.

The fetch goes through a small server-side proxy (browsers can't read other sites directly); it only fetches public pages, caps them at 2 MB, and stores nothing. Pages that build their content with JavaScript after loading may come back thin — for those, open the page, select the content, and use the rich-text paste instead.

HTML table to Markdown

HTML tables become GitHub-flavored Markdown tables, with align and text-align turned into :---, :---:, and ---: column alignment and pipes inside cells escaped. Markdown's table syntax is simpler than HTML's, so a few rules apply:

• A Markdown table always has a header row. If the HTML table has no <thead> or <th> cells (common in Google Docs and Word), the first row is used as the header.
• Line breaks inside a cell become <br>, since a Markdown table row must stay on one line.
• colspan is padded with empty cells; rowspan has no Markdown equivalent.
• Tables containing lists, headings, block quotes, or code blocks are kept as HTML (valid in GitHub Markdown) instead of being mangled; layout tables nested inside other tables are flattened to their content.

Markdown style options explained

• Headings — ATX (# title) works for all six levels; setext underlines (=== / ---) only exist for h1 and h2, so deeper levels still use #.
• Code blocks — fenced blocks keep the language (```python) for syntax highlighting; indented blocks (four spaces) are the original Markdown style. Fences grow automatically when the code itself contains backticks.
• Link style — inline [text](url), or reference style [text][1] with the URLs collected at the end, which reads much better in link-heavy documents.
• Keep images / keep links — turn them off to get text-only Markdown (links keep their text), useful for LLM prompts and word counts.
• Main content only — for whole pages: drops navigation, headers, footers, sidebars, cookie banners, and share widgets before converting.

HTML to Markdown in Python, JavaScript, and pandoc

Converting in bulk or in a pipeline? These are the standard tools.

Python — markdownify(the usual answer to "Python HTML to Markdown"):

# pip install markdownify
from markdownify import markdownify as md

markdown = md(html, heading_style="ATX", bullets="-", strip=["script", "style"])

Python — html2text, older and very configurable:

# pip install html2text
import html2text

h = html2text.HTML2Text()
h.body_width = 0          # don't hard-wrap lines
h.ignore_images = False
markdown = h.handle(html)

JavaScript — Turndown, the library this page uses, in the browser or in Node (with a DOM like jsdom or linkedom). A few lines turn it into your own HTML to Markdown API:

// npm i turndown @joplin/turndown-plugin-gfm
import TurndownService from "turndown";
import { gfm } from "@joplin/turndown-plugin-gfm";

const td = new TurndownService({ headingStyle: "atx", codeBlockStyle: "fenced" });
td.use(gfm);
td.remove(["script", "style"]);

const markdown = td.turndown(html);

Pandoc — from the command line, for files or URLs:

pandoc page.html -f html -t gfm --wrap=none -o page.md
pandoc https://example.com/post -f html -t gfm --wrap=none -o post.md

Frequently asked questions