Turn a web page into plain text or Markdown

This is the fastest tool on the site because it does the least. The server requests the URL the way a plain HTTP client would, parses the HTML, decides which part of the document is the content and which part is furniture, and serialises the content as text or Markdown. No browser is launched and no script on the page runs, which is why a result comes back in well under a second for most pages and why some pages come back nearly empty.

Link to text: what gets kept

The content heuristic scores blocks of the document by text density, link density and position, the approach readability-style extractors have used for fifteen years. Headings, paragraphs, lists, block quotes, tables and code blocks inside the winning region are kept. The Markdown mode preserves their structure with ATX headings, GFM tables and fenced code; the text mode flattens them to paragraphs separated by blank lines. Menus, cookie notices, share bars, related-article rails and footers are dropped. Images are replaced by their alt text where one exists, and by nothing where it does not.

Page to text converter: the token estimate

The number beside the output is characters divided by four and is labelled an estimate on purpose. Tokenisers differ by model, and German text with long compound nouns tokenises worse than English. The estimate is there to catch the difference between a 2,000-token article and a 40,000-token forum thread before either reaches a context window, not to bill against.

What this tool cannot see

A page that renders its content with JavaScript after load, a page behind a login, a page that sends a consent wall instead of the article until a choice is made: all of these return a skeleton, because the server saw what a client without JavaScript sees. The result says so when the extracted content is suspiciously short relative to the HTML, and offers to open the same URL in the agent view, which loads it in a real browser, answers the cookie banner and reads the page the way an agent would. That path costs a browser session and takes a few seconds; this one costs nothing but a fetch.

It also does not transcribe. A link to a video, a podcast or a PDF returns the text that surrounds the media on the page, which is usually a title and a description. The media itself is not downloaded.

Nothing is kept

The fetched HTML and the extracted text exist for the duration of one request. A usage journal records that a fetch happened, when, and how large the response was, as the privacy policy describes; the URL you entered and the text that came back are not written anywhere.