Document tools

PDF to Markdown with an extraction preview

Extract readable PDF text, inspect each page and download a Markdown reference file. The output keeps numbered page sections and marks pages where no text was found. It gives you a reviewable starting point for notes or a knowledge base.

How to convert a PDF to Markdown

1. Choose a readable PDF. Use a file up to 5 MB and 50 pages. Password-protected files need an unencrypted review copy.
2. Inspect the page preview. Choose each page in the selector and compare the extracted text with the original PDF open in your usual viewer.
3. Check gaps and ordering. Review pages with no text, repeated headers, columns, footnotes and table values. Correct the reference text from a verified source before using it.
4. Download or copy. The Markdown contains one numbered section per PDF page. Keep the original filename and version with the exported text.

Try a PDF designed to expose missing text

The two-page fictional example includes a selectable text page and an image-only page. Page 1 names Alex Chen as the reference-review owner and October 15, 2026 as the deadline. It also states that submission is not approval.

On page 1, compare the North and South budget labels with 12000 and 8000. These values form a visual table in the original. The converter extracts text; it does not claim to recover spreadsheet cells or calculate a total. Check that the labels remain associated with the intended values before copying them into another document.

Page 2 contains an image with the code CEDAR-42. There is no text layer on that page. The expected result is a missing-text marker, not the code. This deliberate gap demonstrates why a successful file read is not proof that every visible fact was extracted.

What this Markdown output represents

The converter reads the PDF text items in the order provided by the parser and retains its line endings where available. It adds a heading such as “Page 1” for each page. These headings identify physical page positions in the uploaded file; they are not claims about printed page labels or document sections.

Original text is escaped as literal Markdown. A hash sign in a source sentence should not accidentally become a new heading, and HTML-like text should remain visible text. The preview shows the export directly, so you can inspect it before opening the file in another Markdown application.

The tool does not infer a heading hierarchy from font sizes, summarize passages or supply missing context. You can add meaningful section headings after checking the source. Keep that editorial work separate from the extracted text when it changes the structure or interpretation.

Scanned pages and OCR layers need different checks

A page can look complete in a PDF viewer while containing no selectable text. This is common with scanned pages or screenshots saved as PDF. This converter does not run optical character recognition. A no-text marker may also indicate a blank page or an unreadable text layer, so inspect the original before deciding which recovery step is appropriate.

If you use OCR elsewhere, compare dates, amounts, names and punctuation with the scan. An existing OCR layer can contain recognition errors even though extraction returns plausible sentences. Re-export the corrected PDF or use verified reference text; do not treat the presence of output as an accuracy score.

The ChatGPT PDF guide explains how to check answers against source pages. Asking an AI to convert a PDF does not remove the need to inspect its handling of scans and visual material.

Why columns and tables can read in the wrong order

PDF drawing order and human reading order are not always the same. A page with two columns might contain text items that alternate between them. A footnote might appear near the middle of the extracted passage. Those problems can change the meaning of a sentence even when every individual word is present.

Tables need especially careful review. A label, amount and unit can be drawn separately. Without reliable cell boundaries, a row of numbers is not a usable data model. This tool does not guess the missing structure. For a source workbook, use the Excel and CSV converter to select the actual sheet and range.

When only the PDF is available, compare the affected table with the original and reconstruct a small verified reference table manually in your editor. Include the unit, date and relevant exception. Do not publish a neat-looking table whose cell relationships have not been checked.

Keep the source context with the export

A Markdown file is easier to reuse when it states where the text came from. Add the original filename, document title, version, owner and extraction date. If the PDF is replaced or its page order changes, review the extracted reference and its page citations again.

Repeated headers and footers may appear in the text because they are real text items in the PDF. Remove them only after deciding they add no meaning. A footer containing a version or confidentiality statement may be relevant to how the source should be interpreted or shared.

For a knowledge base, keep the original PDF available alongside any curated text according to your access rules. The starter kit provides a source register and update log. Direct PDF upload is also supported by Ask Your Docs; conversion is an optional preparation workflow.

Limits and processing method

Files are limited to 5 MB, 50 pages and 500,000 extracted characters. Extraction stops after 30 seconds, and you can cancel sooner. A smaller PDF containing the relevant pages is often easier to inspect than an entire archive.

The tool uses Mozilla PDF.js with its browser worker. It reads the selected file’s text layer and produces a local Markdown download. It does not execute PDF JavaScript, contact an AI model, run OCR or upload the selected document to our server.

Clearing the file or leaving the page releases the active conversion state. Files you download and text you copy remain in the destination you choose. Review both the contents and the intended audience before sharing an exported reference.

Frequently asked questions