Extract text from a PDF

How to extract text from a PDF

1. Drop the PDF onto the box above. It is parsed by pdf.js in this tab; nothing is uploaded.
2. Wait for the progress bar — extraction runs page by page and stays responsive even on long documents.
3. Review the per-page output. Each page has its own copy button so you can take just one section.
4. Toggle the page separators on or off depending on whether you want --- Page 3 --- markers in the file.
5. Copy everything, or download a UTF-8 .txt file.

Two kinds of PDF, and why it matters

Every PDF falls into one of two categories, and the difference decides whether text extraction is possible at all.

A quick check before you start: open the PDF in any reader and try to select a word with the mouse. If the selection highlights letters, this tool will give you clean text. If it draws a rectangle over the whole page, the page is an image.

How the reading order is reconstructed

This is the part most extractors get wrong, so it is worth explaining. A PDF does not store paragraphs, sentences or even lines. It stores instructions like "draw these glyphs at this position with this font". pdf.js hands back those positioned runs in whatever order the generator wrote them, which is frequently not reading order.

To turn that back into prose, the tool:

• groups glyph runs into lines by comparing their baselines, with a tolerance scaled to the font size;
• sorts the lines from the top of the page downwards, and each line's fragments from left to right;
• inserts a space where two fragments are further apart than the font's natural letter spacing, which recovers word gaps that the PDF stored as positioning rather than space characters;
• measures the typical gap between lines, then starts a new paragraph whenever the gap is noticeably larger, the line is indented, or it begins with a bullet or a number;
• re-joins words split across a line break by a hyphen.

The honest caveat: sorting by vertical position is correct for single-column pages and wrong for magazines, newspapers and academic two-column layouts, where it interleaves the columns. Tables come out as rows of words without their column structure. No purely geometric approach avoids this, and any tool that claims otherwise is guessing too.

Useful things to do with the output

• Search and quote a long report without scrolling through a viewer.
• Count words in a manuscript or a submission with a defined limit.
• Feed a language model or script that expects plain text rather than a binary format.
• Diff two versions of a contract by extracting both and comparing the text.
• Rescue content from a PDF whose source document has been lost.

If you want an editable document rather than plain text, the PDF to Word converter wraps this same extraction in a real .docx. To go the other way, see Text to PDF.