PDF to Markdown

PDF is where documents go to stop changing. That is its purpose and its virtue, and it is exactly the problem when the content needs to live on — as a page on a documentation site, an entry in a knowledge base, a file in a repository, or the input to something that processes text.

Markdown is the opposite kind of format: plain text, small, diffable, readable without any particular software, and convertible into almost anything else. Moving content from PDF into Markdown turns a finished artefact back into something that can be edited, reviewed line by line, and republished.

The common destinations are static site generators, wikis and documentation platforms, note systems, and version-controlled repositories where a meaningful diff matters more than presentation.

A PDF stores glyphs at coordinates. It does not record that a line was a heading, that three lines were one paragraph, or that a column of items was a list — those are things a reader infers from appearance. A converter has to make the same inferences from the same evidence.

Type size carries most of the signal. Text substantially larger than the body is almost always a heading, and the distinct sizes present in a document map onto heading levels in order. Bullet characters and leading numbers identify lists. Line spacing and indentation reveal where paragraphs begin and end, which is what allows wrapped lines to be rejoined into flowing text rather than preserved as hard breaks.

This works well because most documents are typographically conventional. It works badly on documents that are not — where size means emphasis rather than hierarchy, or where a designer has made body text large and headings small.

Single-column, text-led documents are the strong case: reports, papers, policies, documentation, ebooks, meeting notes. Headings, paragraphs, lists and emphasis all come through, and the result is usable with light editing.

Three things reliably need attention. Multi-column layouts force a reading-order decision that is sometimes unresolvable, so an academic paper or a newsletter may interleave columns. Tables rarely survive as tables, because a grid of aligned text has no signature distinguishing it from four columns of separate text unless it was drawn with rules. And headers, footers and page numbers repeat on every page and land in the flow, since they were page furniture rather than content.

None of that is a defect in the conversion so much as a consequence of what a PDF is. Budgeting a few minutes to delete repeated furniture and rebuild a table or two is the realistic expectation.

Check first whether the PDF has text at all: try selecting a word. If nothing highlights, the page is an image and OCR PDF has to come first — running it turns a hopeless conversion into a good one.

Then check the output top to bottom in an editor with a live preview, which makes wrong heading levels obvious immediately. Fixing the heading hierarchy first is worth doing before anything else, because it determines how the document reads everywhere it is published.

If you need the images as well as the words, PDF to JPG in extract mode pulls them out at native size. And if what you actually wanted was an editable document rather than a text format, PDF to Word reconstructs paragraphs and real tables and is the better tool.