PDF to Text

Why copying from a PDF reader goes wrong

Selecting text in a PDF viewer and pasting it usually produces a mess: line breaks in the middle of sentences, hyphenated words split across lines, headers and footers interleaved with the body, and columns run together. That is not the reader being careless — it is the honest consequence of what a PDF is.

A PDF does not store paragraphs. It stores instructions to place particular glyphs at particular coordinates. Any tool recovering text has to infer, from position alone, which characters belong to the same word, which words to the same line, and which lines to the same paragraph. Different tools infer differently, which is why the same document extracts differently everywhere.

This tool uses pdf.js, the engine Firefox uses to render PDFs, so the text it recovers is exactly what the renderer sees. The layout options let you choose which inference you want rather than accepting one silently.

Choosing a layout mode

Flowing prose joins everything into continuous text with single spaces. This is right when you want to read or edit the content and the original line breaks are meaningless — which is true of most narrative documents, where a line break is just where the typesetter ran out of width.

Preserving line breaks groups characters by their vertical position, so each visual line becomes a line of text. This is what you want for tables, code listings, addresses, poetry, or anything where the arrangement carries meaning. It is also better when you intend to clean the result up by hand.

The third option adds a marker between pages. Use it when you need to cite page numbers, or when you are going to split the output back into per-page chunks later.

When there is no text to extract

If the tool reports that no text was found, the PDF is almost certainly a scan. A scanned document is a photograph of a page wrapped in a PDF container: to a computer it is pixels, with no characters anywhere in the file. This is why you cannot select text in it in any reader either.

Recovering words from a scan requires optical character recognition, which examines the image and guesses at the letters. That is a fundamentally different operation from extraction, it is never perfectly accurate, and this tool does not perform it.

A useful diagnostic: open the PDF in any reader and try to select a word. If the selection highlights a rectangle rather than the letters, or nothing highlights at all, it is a scan and no extraction tool will help.

A mixed document — some pages typed, some scanned — will extract the typed pages and report the scanned ones as empty, which the summary tells you.

What extraction loses

Formatting, entirely. Bold, italics, font sizes, colours, indentation and alignment are all presentation, and plain text has no way to carry them. Tables lose their structure and become sequences of cell contents, which is why a table extracted as flowing prose is usually unreadable and worth extracting with line breaks preserved instead.

Images, charts and any text baked into them are not extracted, because they are not text. Footnotes usually appear at the point where they physically sit on the page rather than attached to their reference.

Reading order in multi-column layouts is the weakest point of any extraction. A two-column academic paper often extracts with the columns interleaved line by line, because that is genuinely the order the characters appear in the file. If you need clean text from a columned document, extracting page by page and repairing by hand is usually faster than fighting it.