HTML to Markdown

There are two ways to feed this tool. Select part of a web page, copy it, and paste into the left box: the browser hands over the formatted content as HTML behind the scenes, and the tool picks that up instead of the plain text. Or paste raw HTML source, for example from view-source or a saved file. Either way the Markdown appears on the right with a copy button, and headings, lists, links, code and tables are all preserved.

The most common reason to do this is to hand web content to ChatGPT, Claude or Gemini. A page copied straight from the browser carries navigation, inline styles, tracking scripts and dozens of nested tags that a language model has no use for, and every one of those characters costs context window and money. Markdown keeps the structure the model actually reads while dropping the wrapping, so the same article usually shrinks to a fraction of the tokens. The line under the result shows the before and after counts using the o200k_base tokenizer that GPT-4o and GPT-4.1 use; other models will differ a little, but the ratio is representative.

Two options trim further. Remove images drops image tags entirely, since a model reading text cannot see them and the URLs only add noise. Remove links keeps the link text but discards the URL, which matters on pages where every other phrase is a link. Scripts, styles, inline SVG icons, comments and the document head are always removed. HTML tables become GitHub-style pipe tables; a table nested inside another is flattened to text because Markdown has no way to express it.

Everything runs in your browser. What you paste is read from the clipboard event and never sent anywhere, so internal wiki pages or documents behind a login are safe to convert. Very long pages take a moment because the tokenizer has to walk the whole text twice; inputs over a million characters are refused, so split those first.