Try to select a sentence in some PDFs and the cursor turns into a crosshair, dragging a rectangle over the whole page. The document looks like typeset text. It contains none.
Text is drawn, not stored as sentences
A PDF is a series of drawing instructions. For text, the page content stream says roughly “place glyph 47 of font F3 at these coordinates, then glyph 12, then glyph 103.” The viewer maps glyph numbers to character codes through a font’s encoding table, and ideally through a ToUnicode map that records exactly which character each glyph represents. Copying works only when that mapping is present and correct.
Break any link and you get one of these:
- Nothing selectable — the page is a scanned image. There are pixels, not glyphs.
- Gibberish on paste — the font has glyphs but a broken or missing
ToUnicodemap. You copy a trail of symbols where plain letters should be. - Outlined text — a design tool exported letter shapes as vector paths, usually to preserve a licensed font. The glyphs are literally drawings now.
- Copy blocked — the file is encrypted with permissions that disallow copying. Many viewers can ignore this; a viewer that does not will refuse.
Confirming which case you have
Click once on a paragraph and press Ctrl+A. If the highlight snaps to text lines, a text layer exists and the problem is encoding. If it covers the page as one block, it is a scan.
Getting the text anyway
For text-layer PDFs, a different extractor often copes better. The PDF to text tool runs pdf.js in your browser, walks every page’s text content, and dumps the text runs into a textarea — pages separated by a blank line, runs joined with spaces. Nothing is uploaded, and the result can be copied or downloaded as a .txt file. What it cannot do is invent text: feed it a scan and the output is empty or a few stray characters.
For scans and outlined text, the only route is OCR — converting pictures of letters back into characters. Expect to proofread the result; OCR output is a draft.
The rule: if you cannot select a single character, stop clicking. There is no text layer, and no extractor will create one.