PDFGym
EN

Turn a PDF into Markdown

or drop one here
Choose a PDF first.

Documentation trapped in PDF is documentation nobody updates. This page reads the selectable text, rebuilds it in reading order with headings, lists, and code blocks, and verifies that 98 percent of the kept characters survived in sequence. Scanned pages have no text to free, so those files are refused with the reason named.

The manual that had to become wiki pages

When the source file is gone and only the PDF remains, retyping sixty pages is not a plan. Extract the text with its structure and clean that instead.

A tech writer inherits field-manual-2024.pdf, 60 pages, no source, and a directive to move it onto the company wiki. She opens it here, keeps page markers on so every section stays traceable to its page, leaves repeat removal on to drop the running heads, and presses Convert to Markdown.

The download lands as field-manual-2024.md with headings inferred, the proof line counting inspected pages and retained text. She pastes it chapter by chapter into the wiki, markers showing where each PDF page began. An afternoon of cleanup beats a month of retyping.

Four passes from PDF to Markdown

  1. Open the PDF

    Press Choose a PDF, or drop the file on the page. The document needs selectable text to give.

  2. Set markers and repeats

    Pick Keep page markers or Continuous document, then Remove repeats or Keep every line. A note under the options restates the combination before you run.

  3. Convert the text

    Press Convert to Markdown. Lines are rebuilt in reading order, and font size and style evidence infers headings, lists, and code blocks.

  4. Check and download

    The Markdown is read back: marker count must equal page count, characters must survive in order above the retention floor, and encoding errors are refused. Download Markdown then saves the file under the source name, beside a retention percentage and notes on removed repeats and textless pages.

What the converter infers and what it proves

A PDF records positioned runs; Markdown records ordered structure. The tool groups runs into lines, lines into reading order, and promotes the visually loud ones to headings, bullets to lists, and monospaced runs to code. Page markers arrive as HTML comments naming each page, invisible once rendered, and repeat removal drops only short lines repeating at the same page edge across most of the document. The result note tells you exactly how many lines went.

The proof then guards three things: every page marker present and counted, the visible characters surviving in their original sequence, and retention at or above 98 percent with no encoding replacement characters. The preview shows the source page beside the first chunk of Markdown as plain text, deliberately unrendered so nothing executes. Skim it before trusting the structure.

Documents that convert badly

Scans refuse outright: no text layer, no Markdown, with OCR named as the missing step. Guessed text needs its own accuracy measure and human review, so the tool declines to invent it. Multi-column papers, dense tables, and footnotes convert approximately, with reading order doing its best and the preview showing where it guessed.

Repeat removal has one sharp edge. A coupon code, warning, or slogan printed at the same edge of every page looks exactly like a running head to the rule, and it will be dropped with the count noted. When repeated lines are content rather than furniture, rerun with Keep every line and delete the true headers by hand.

When Markdown is the wrong target

An editable office document comes from PDF to Word, and a visual page copy from PDF to HTML. Tables with real grids belong to PDF to CSV or PDF to Excel. To round-trip back into print, Markdown to PDF renders Markdown into a verified document again. Page pictures for slides come from PDF to images.

Why drafts stay on your device

Unpublished research, legal drafts, and internal reports should not audition on a stranger server. Extraction, inference, and the retention check all run where the PDF already sits, with no account and nothing retained.

The failure mode is silence, not garbage: below the retention floor, or with encoding gremlins aboard, there is simply no download. A missing file tells you the document needs a different route, which beats a confident file with holes.

Short answers about PDF to Markdown

Why does my two-column paper read across instead of down?
Reading order follows visual lines, and two columns look like one wide line to the grouper. The words all survive, but column papers need manual reordering after download.
A coupon code repeats on every page and vanished. Where is it?
Repeat removal ate it: a short line at the same edge across most pages matches the running-head rule. Rerun with Keep every line to bring it back.
What does 98 percent retention actually enforce?
That the visible characters carried over in their original sequence, nearly all of them. Below the floor, or with encoding replacement characters aboard, the download is withheld instead of shipping holes.
Can I trust the inferred headings?
Mostly. Bigger, bolder lines earn the marks, and the proof line counts how many were inferred. Skim the preview anyway; a loud caption can pass as a heading.
The preview shows raw # marks. Is that broken?
No. The preview is plain text on purpose: it never renders HTML, follows links, or loads images. Paste the download into any Markdown app to see it dressed.

More tools that move words around

The tools below treat text the same careful way: extracted on your device, checked before download, and never asking for an account.