Read the selected pages
All pages are selected by default. Enter page positions such as 2, 6-8 to reduce the conversion scope. Positions count from the start of the file, even when printed page labels differ. The full PDF must still open.
How PDF to Markdown conversion works
Convert the pages you need, then compare the draft with the PDF. This guide explains the page options, what the tool can preserve, and what our tests establish.
From selected pages to a draft
Recognize English scans
When a selected page has image content but no readable text layer, the tool automatically tries English text recognition. Recognition resources load at that point. Pages with both images and existing text need source review; automatic recognition does not replace that text.
Review the result
The result identifies included pages and pages without a selected text draft. Open page options to compare the source, choose an alternative table draft, adjust a column layout or retry recognition. Native text is included as a draft even when a page has layout warnings. Table and column alternatives remain selectable. Only usable, selected drafts appear in the Markdown.
What to expect from each kind of content
Native text and formatting
Native text can retain identifiers and link references. Suggested headings, lists and code formatting remain reviewable. Columns, raised numbers, labels and hidden text layers can have the wrong order or words even when extraction finishes.
Tables and columns
Eligible native-text pages can offer bordered-table Markdown. The first row becomes the header; line breaks in cells use <br>. Merged cells and automatic joining across pages are unsupported. Manual column and region settings provide alternative drafts; compare them with the source.
Recognized text and non-text content
English recognition provides a text draft. It can miss or misread words and does not reproduce pictures, original layout, equations or table relationships. Image and annotation signals produce page notes; those signals do not identify every possible omission.
How to interpret the result
Text in the draft
A page count reports how many needed pages contribute text. It does not report a percentage of correct content. Initial page reading and scan recognition are separate processing stages.
Missing pages and page notes
A page may need a draft choice, have no recognized words, reach a resource limit or fail recognition. Its note gives the next action. Other usable pages can still be downloaded with warnings. A source or output integrity error can pause delivery.
Downloads and the report
A usable draft is available to download or copy without a review checkbox. These actions do not record source comparison or certify accuracy. Changing the source, pages or output settings clears old links; saved files keep their existing content. The conversion report records scope, output choices and warnings.
Resource limits and cancellation
- One PDF at a time, up to 50 MiB. Page selection cannot bypass a password or repair a damaged file.
- There is no fixed selected-page count limit. Inspection budgets are 100,000 retained text items, 500,000 operations and 2,000,000 characters per selection; one page is limited to 25,000 text items and 100,000 operations. These are safeguards, not measurements of free memory.
- English recognition runs sequentially, one page at a time. Each page is bounded to 4 million rendered pixels, 90 seconds and 200,000 output characters.
- A recognition failure stops later automatic recognition attempts in that run. Page options allow retry. Cancellation stops active work and clears review; already-started resource requests may still finish.
- The full file must open, and large images use device memory. A smaller selected range can reduce conversion work but cannot guarantee that a particular device can handle the source.
Public test files and what they establish
These authored English examples are controlled checks, not an accuracy rate for arbitrary PDFs. Download a source and its actual output to compare the observations yourself.
| Source and route | Actual output | Previously observed result |
|---|---|---|
| One-page text PDF · native text | Markdown · Report | 00123 and the two example link targets retained. |
| Bordered billing table · reviewed table choice | Markdown · Report | Header and four data rows matched expected cells, including leading zeros and displayed amounts. |
| English scan · automatic recognition | Markdown · Report | Heading, 123.45 USD and closing sentence retained; warnings record unsupported image preservation. |
Prior controlled checks also exercised page selection, long plain-text documents, manual column layouts, resource-load failure, timeout, cancellation and retry. Known counterexamples include misplaced footnotes, mixed footer/body order and a skewed low-resolution scan returning no recognized words. See the homepage test records for dates and distinctions between browser-window checks and owner-reported phone checks.
Local handling
The PDF, previews and output stay in the browser session. Conversion resources are served by this site; loading them does not upload the PDF. Website delivery and analytics are separate; see Privacy. Return to the converter.