High-quality PDF to JSON converter

Turn a PDF into structured JSON.

Extracting text from a PDF gives you lines, not a document. WandPDF rebuilds the structure and writes it as JSON: every heading, paragraph, list, table and footnote as its own block, in reading order.

Convert to
Tested on 31 books: 25 came out clean with WandPDF, 2 with Calibre (CloudConvert's engine). See the test 

PDF to JSON, before and after

Same page. Two very different extractions.

A typical converter dumps the page text line by line. WandPDF writes the structure.

Formats

Convert PDF to EPUB, Kindle, Markdown, text, HTML or JSON.

.epub

PDF to EPUB

An EPUB ebook for Apple Books, Kobo, Google Play Books and any reader app. Chapters become a table of contents you can tap.

.azw3

PDF to Kindle (AZW3)

An AZW3 file for Kindle devices and apps. Change the font size and the text flows, instead of zooming a tiny page.

.md

PDF to Markdown

Clean text with headings, lists and tables for notes, Obsidian, a wiki or as input for ChatGPT, Claude and other LLMs.

.txt

PDF to TXT

Plain text with whole paragraphs, no page numbers or broken lines. Scanned PDFs are read with OCR.

.html

PDF to HTML

One readable web page with real headings, tables and footnotes, pictures inside. Reflows on phones, not a copy of the printed page.

.json

PDF to JSON

The document as typed blocks in reading order: headings, paragraphs, lists, table cells, footnotes. For RAG, search and scripts.

NextWord is next: an editable document, not a copy of the page layout.

How it works

How to convert a PDF to JSON

  1. 1

    Drop your PDF

    Any PDF: a paper, a report, a manual, a book, or a scan of printed pages.

  2. 2

    Pick JSON

    One .json file: the title, the contents and a flat list of typed blocks.

  3. 3

    Use the blocks

    Chunk on headings for RAG, index paragraphs for search, or read tables cell by cell in a script.

Under the hood

What a PDF hides, and WandPDF fixes.

Tested on 31 books the parser was never tuned on.

Clean means no major defects such as broken paragraphs, stray page numbers, running headers or lost text.

Full test with Marker and Docling →
WandPDF25 / 31 clean
Calibre 8.5 (used by CloudConvert)2 / 31 clean

FAQ

PDF to JSON FAQ

What does the JSON look like?

An object with title, author, lang, pages (the page count), toc and blocks. Each block has a type (heading, paragraph, list, table, figure, math, code, footnote, quote or note), the chapter it belongs to and its content: text for most, items for a list, rows of cells for a table, LaTeX for a display formula. Each block also has page and bbox fields, described below.

Is it good for RAG and LLM pipelines?

That is what it is for. Running headers, printed page numbers and hyphenation are removed from the text, paragraphs are whole and in reading order, and headings are marked, so you can split on real sections instead of fixed-size windows. Every block also keeps its PDF page, which lets an answer cite its source.

Does each block come with its page and bounding box?

Yes. Every block has a page field, the PDF page it starts on (counting from 1), and a bbox field, its bounding box on that page as [x0, top, x1, bottom] in points (1/72 of an inch), measured from the top-left corner of the page with y growing downward, as in pdfplumber. Use them to cite the source page or highlight the passage in a PDF viewer. The page field is the page's position in the file, which may differ from the number printed on it. A block that runs over a page break or across columns also has its own pages and bboxes arrays, with one entry per part.

How are tables written?

As rows of cells, each with its text and, where they apply, a header flag and its colspan and rowspan. Tables drawn with ruled lines come out most reliably; some tables without any lines still come out as paragraphs.

Are the pictures in the JSON?

No. A figure block carries the picture's alt text and caption, not the image itself. Pick EPUB, HTML or Markdown to keep the pictures.

Does it work with scanned PDFs?

Yes, in English, German, French, Spanish, Italian, Portuguese and Russian. When the pages are pictures, WandPDF recognises the text on each one (OCR) and builds the book from it, so you can resize the type and search it. A long scanned book takes several minutes, and the page shows how far it has got.

Which languages work?

English and other Latin-script languages, plus Chinese and Japanese. Arabic, Hebrew and other right-to-left scripts are still in progress.

What happens to my file?

It is kept only long enough to convert it and let you download the result, then deleted.

Your next PDF deserves clean data.

EPUB, Kindle, Markdown, text, HTML and JSON from one upload.

Convert a PDF