Use a tool that reads the PDF text layer and rejoins lines that the layout wrapped. Copy and paste fails because a PDF stores text as positioned fragments, so pasted output arrives broken at every visual line end.
Selecting a few paragraphs in a PDF and pasting them elsewhere produces text broken at the end of every visual line, with hyphens stranded mid-word and headers and page numbers scattered through it. For more than a paragraph or two, it is unusable.
Why copy and paste produces a mess
A PDF does not store paragraphs. It stores instructions to draw text at coordinates — often one fragment per word, sometimes per character, in whatever order suited the generator. The visual line breaks you see are a consequence of where things were placed, not markers in the content.
When you copy, you get those fragments in roughly positional order with a line break at each visual line end. The paragraph structure was never in the file to be copied.
What a proper extraction does
A dedicated tool reads the same text layer but reconstructs from it, which is where the difference lies:
Rejoins wrapped lines into paragraphs, using line spacing and indentation to work out where a paragraph really ends.
Reinstates spaces between words, which PDFs frequently omit because each word was positioned separately rather than separated by a space character.
Removes running headers and footers repeated on most pages — the page furniture that otherwise appears every page or so in your output.
Open the PDF and try to select a sentence. If nothing highlights, the file contains no text — it is a scan, a photograph of a page.
No extraction tool can help, because there is nothing to read. Getting text out requires optical character recognition, which examines the image and recognises letter shapes. That is a genuinely different technology, and not every tool includes it — this one does not.
Why columns come out interleaved
Two-column documents — academic papers, newsletters, some reports — often extract as alternating fragments from each column. This is not a bug so much as an unanswerable question.
The file records glyphs at coordinates. Reading left to right across the page is a perfectly reasonable interpretation; so is reading down one column then the other. Nothing in the file says which is correct, so extraction picks one and sometimes picks wrong.
The practical workaround is to extract a page at a time and reassemble, or to select one column at a time in a viewer where you can see what you are selecting.
Choosing the right output
What you plan to do with the text should decide the format.
Plain text — for quoting, searching, feeding into another tool, or reading without formatting. The most reliable option, because there is nothing to go wrong.
Images — when you want the page as it looks rather than what it says. Rasterising pages is the right tool for that.
If you only need the words, take plain text. Converting to Word to get at the text and then stripping the formatting is a longer route to a worse result.
Practical tips
Extract only the pages you need
For a few pages out of a long report, extract those pages first. Less output means less to clean up.
Watch for hyphenation
Justified text often breaks words across lines. Rejoining lines can leave a hyphen mid-word — worth a find-and-replace pass if the document was typeset that way.
Check the beginning and the end
Extraction problems cluster at page boundaries and around headers. If the first and last paragraphs of a few pages read correctly, the middle almost certainly did too.
A note on where this runs
Text extraction is a natural fit for the browser: the PDF is parsed in the tab and the text never leaves your device. That matters more than it sounds, because the documents people extract from — contracts, statements, reports — are often the ones worth not uploading. See is it safe to use online PDF converters? for how to verify a tool actually works that way.