PDF to Markdown Converter - Turn a PDF into Editable .md Text

Convert PDF to Markdown for notes, docs and AI tools

Markdown is plain text with a few simple marks for headings and paragraphs. Notes apps, GitHub, documentation sites and AI chat tools all read it without fuss, and it is easy to search, diff and edit. A PDF, by contrast, is a fixed picture of a page. This tool converts a PDF to Markdown by reading the text off each page, marking the lines that look like headings, and giving you a single .md file to download.

It is a first-draft converter, not a layout engine, and the sections below say plainly what it does well and what it leaves for you to tidy. Think of it as the quickest way to get the words of a PDF into a text editor, with a rough structure already in place, rather than a one-click way to rebuild a formatted document.

How to convert a PDF to a Markdown file

1. 1
   Drop a PDF onto the upload area or click it to browse. You can add several; each one is converted in turn.
2. 2
   Look at the line under the upload box. It tells you how many pages the file has and whether they carry selectable text. A typed document is described as a good fit; a scan gets a warning that OCR will be used.
3. 3
   Leave Read scanned pages with OCR on Only if the PDF has no text unless you have a reason to change it. The other choices are Always - re-read every page and Never.
4. 4
   If the document is scanned or mixed, pick the OCR language: English, Hindi or Hindi + English.
5. 5
   Press Convert to Markdown. A progress percentage shows while pages are read.
6. 6
   Read the preview of the first part of the result, check the note beneath it, and download the converted.md file.

What the converter really does

For a PDF with a text layer, the tool opens the file with pdf.js and asks each page for its text items together with their positions. Items that sit at the same height on the page are joined into one line, and the lines are sorted from top to bottom. This takes milliseconds, because no recognition is involved.

Each page then becomes a block of Markdown. The page starts with a first-level heading such as Page 3, every line is written as its own paragraph, and pages are separated by a horizontal rule (three hyphens). A line is marked as a second-level heading when it is shorter than 70 characters, does not end in a full stop, question mark, exclamation mark, comma, semicolon or colon, and begins with a capital letter or a digit. Everything else is left as plain text.

That rule is simple on purpose. The text of a PDF does not say which lines are headings; it only says where letters sit. Size and weight are not consulted, so the tool guesses from shape alone, and you should expect to correct some guesses.

Worked example: how seven lines are classified

Here is a single typed page with seven lines, run through the converter. The lengths are real character counts, and the verdict applies the rule above.

Five of the seven lines became headings and three of them were meant to be. The two false ones show the weakness of the rule: a short table label and a sentence that happens to lack a final full stop both look like titles. Real documents behave the same way, so a quick pass through the headings is part of the job.

For a rough sense of scale when you plan to paste the result into an AI tool, a 30-page report with about 400 words a page holds 12,000 words. At the common rule of thumb of about three quarters of a word per token, that is around 16,000 tokens. Treat the figure as an estimate, because tokenisers differ.

Converting scanned PDFs with OCR

A scan has no text to read, only pictures of text, so the tool uses OCR. It renders each needed page at 200 DPI and recognises it with Tesseract, an open-source engine that runs in your browser as WebAssembly. You can pick English, Hindi or Hindi + English; the last suits bilingual circulars and notices.

The first OCR run downloads the engine and the language data, so it needs an internet connection and takes longer; later runs reuse what the browser has stored. After that, expect a few seconds per page. The result note tells you how many pages were read and the average confidence. Below 70 per cent it warns you to check against the original, and a sharper scan reads far better than any setting can compensate for.

Mixed files need one decision from you. If a PDF has a text layer but some pages are scans, the default mode reads the typed text directly, skips the empty scanned pages and tells you how many it skipped. That keeps a long report with a scanned annexure fast, but the annexure is missing from the result. Switch the setting to Always - re-read every page to have all of them recognised, accepting that the typed pages are then read by OCR as well.

Reading and cleaning the output

• Page headings and rules. Every page gets a first-level heading and a rule between pages. If you want continuous text, search for those markers and delete them.
• Broken paragraphs. A paragraph that wrapped over four lines in the PDF arrives as four short paragraphs. In an editor with regular-expression search, finding a blank line followed by a lowercase letter and replacing it with a space rejoins most of them.
• Wrong headings. Table labels, page numbers (a lone number begins with a digit) and half-sentences can appear with the second-level mark. Delete the marks that do not belong.
• Missing headings. A title ending in a colon or a full stop will not be marked, so add the mark by hand.
• Running headers and footers. Text repeated at the top or bottom of each page, such as a document title, is not removed.
• Hyphenated words. A word split at the end of a line stays split.

What gets left out

Markdown can describe tables, bold text, links, lists and images, but this converter does not reconstruct them. Pictures and drawings are skipped, because only text is read. A table row comes out as one line with its cells separated by spaces, with no pipes or columns, so if you need the numbers in a grid, use extract tables instead. Bold, italic, font sizes and hyperlinks are not carried across, and a bulleted list keeps its bullet characters as ordinary text.

Pages laid out in two columns can read across the gutter, since lines are formed by height on the page. Check such pages first. If the formatting matters more than plain text, an editable Word file is usually the better target.

Where a Markdown version helps

• Study notes: turn a chapter PDF into text you can annotate in Obsidian, Notion or any editor.
• Documentation: start a README or help page from a specification someone sent as PDF.
• AI assistants: paste the text of a long document into a chat or a prompt, where plain text is cheaper and cleaner than a file upload.
• Court judgments, circulars and gazette notices that exist as scans: read them with Hindi or English OCR, then search the result.
• Quoting: copy a paragraph with its heading structure into a blog post or report.
• Version tracking: keep a text copy of a policy in Git and see what changed between versions with the usual diff tools.

Practical tips

• Start with a typed PDF if you have a choice. A text layer is exact, while OCR is a best guess.
• For a very long PDF, split the report into chapters and convert each. Each output is smaller to review.
• If the source is a pile of photographs, first build a PDF from them, then run OCR on it. Clear, straight, well-lit photos read much better.
• Open the .md file in an editor with a Markdown preview so you can see the headings as they will look.
• Compare a few lines of the output against the PDF before you rely on it, especially numbers and names.

Common mistakes

• Expecting the output to match the PDF's appearance. Markdown records structure, not layout.
• Running OCR on a typed PDF. It is slower and usually less accurate than reading the text directly, which is why the default is to read text first.
• Choosing English for a Hindi document. Pick the language that matches the page, or Hindi + English for mixed content.
• Trusting OCR numbers without checking. A misread digit looks as confident as a correct one.
• Trying to convert a password-protected file without opening it first. The page asks you to remove the password before it will run.

Limits and privacy

Each PDF can be up to 100 MB and 500 pages, and the page tells you as soon as a file exceeds that. A damaged PDF is turned away with a pointer to repair the file. If the file has no readable text and OCR is set to Never, you get a message instead of an empty document. The preview on the page shows only the start of the result; the downloaded file holds all of it.

Text extraction and OCR both run in your browser, so the PDF is not uploaded to a Jobsixo server. The one thing fetched from the network is the OCR engine and language data, on first use. Pages are read on your device and the Markdown is created there too.