Pull the text out of a PDF instead of typing it again

Extracting text is the fastest useful thing you can do with a PDF, and it is also the operation people most often replace with twenty minutes of retyping, because the first attempt produced something that looked scrambled.

What extraction actually reads

A native PDF stores text as a sequence of show-text operators: place these character codes, using this font, at this position. Extraction walks that sequence, maps each code back to a Unicode character through the font's encoding tables, and writes out the result.

That means native extraction is exact. It is not guessing at shapes the way recognition does. If the file says the character is a capital B, you get a capital B, every time.

The text extraction tool does this in the browser, which matters when the document is a contract, a statement, or a report you would rather not upload just to get a paragraph out of it.

Check first whether there is any text at all

Try to select a word in your reader. If individual characters highlight, the text is there. If a whole rectangle highlights, or nothing does, the page is an image and there is nothing to extract.

For that case, run optical character recognition first to add a text layer, then extract from the recognised file. The OCR guide covers how to get an accurate layer out of an imperfect scan.

Why the output sometimes looks scrambled

Extraction follows the order in which text was drawn, and the drawing order is whatever the producing application chose. It has no obligation to match how a human reads the page. That single fact explains nearly every disappointing result:

• Multi-column layouts can interleave, giving you a line from column one followed by a line from column two, if the generator drew across the page rather than down each column.
• Headers, footers, and page numbers appear inline in the middle of the flow, repeating every page.
• Sidebars, pull quotes, and captions land wherever they were drawn, often interrupting a sentence.
• Tables flatten into a stream of cell values with no row structure, because the row structure was never in the file.
• Footnotes arrive detached from the text that referenced them.

None of this is a malfunction. It is the honest result of a format that stores positions rather than a reading sequence.

Cleanup you will almost always need

Budget a couple of minutes for these, and the text becomes genuinely usable:

1. Soft hyphens. Words broken across line ends often extract as manage- and ment. Search for a hyphen followed by a line break and rejoin.
2. Ligatures. The single glyph fi can extract as one character that does not match a search for "fi". The same applies to fl, ff, and ffi.
3. Repeated furniture. Delete the running header and footer that now appears once per page.
4. Line breaks inside paragraphs. Extraction preserves the visual line breaks, not the paragraph structure. Rejoin lines that do not end a sentence.
5. Quotation marks and dashes. Typographic quotes and em dashes may need normalising if the text is going into code or a data pipeline.
6. Spacing. Some generators emit no space characters at all and rely on positioning, which produces runtogetherwords needing manual separation.

If the output contains broken lines, repeated headers, interleaved columns, or table fragments, use the repeatable cleanup sequence in pull clean text out of a messy document instead of correcting each page ad hoc.

When to reach for a different tool

Extraction gives you words, not layout. If what you actually want is an editable document with headings, styles, and structure intact, the PDF to Word converter is the closer fit, and the conversion guide explains how much structure realistically survives.

Use plain extraction when you want the words themselves: quoting a clause, searching a document set, feeding text into a script, checking what a page really contains beneath its images, or verifying that a redaction removed what it was supposed to remove. For all of those, unstructured text is not a limitation — it is precisely the point.