Convert HTML to Markdown in your browser

Markdown is a smaller language than HTML, so every converter is a set of decisions about what to keep. This one is tuned for one purpose: producing text that a model reads well and a person can still edit. That biases it towards structure and away from presentation.

Kept

Headings h1 to h6 become ATX headings. Paragraphs, ordered and unordered lists with nesting, block quotes, horizontal rules, links with their href, images with their alt text and src, emphasis and strong, inline code, and line breaks inside a paragraph all have a direct equivalent and get it. Tables become GFM tables, with the header row taken from thead or from the first row of th cells, and cell content flattened to one line because Markdown tables cannot hold a paragraph. Fenced code blocks carry the language when the class on the code element names one, in the language-x or lang-x convention.

Dropped

script, style, noscript, template, iframe and object are removed with their content. Comments are removed. Inline styles, classes, ids and data attributes have no representation in Markdown and disappear, which is most of the point: a page's HTML is frequently ten times the size of its content, and nearly all of the difference is markup a model does not need. Empty elements and runs of whitespace collapse.

Cannot know

The converter does not fetch anything. Relative links stay relative, because it does not know the base URL unless the document carries a base element. Images referenced by URL stay references. Content that a script would have inserted is not there, because no script runs. And the converter has no opinion about which part of a page is the article: if you paste a whole page, you get the whole page in Markdown, navigation and footer included. The website-to-text tool is the one that makes that judgement, for a URL rather than pasted markup.

In your browser

The conversion is a pure function over the pasted string and runs entirely in the page. No request carries the HTML, and no request carries the Markdown; you can confirm that in the network panel. Nothing is stored, and closing the tab discards the input. This is also why the tool works offline once the page has loaded, and why there is no size limit beyond what your browser tab is willing to hold.

The token count is characters divided by four, and the label says estimate because it is one. Tokenisers vary between models, Markdown syntax characters tokenise differently from prose, and tables in particular cost more tokens than their text suggests because every pipe and dash is a token. Use the number to compare two versions of the same document, not to bill against.