You scan a stack of receipts, press Ctrl+F to find one invoice number, and nothing turns up. That is because a scanned page holds no text at all, only pixels that happen to look like letters.
What OCR adds
Optical character recognition looks at each page image, works out which characters are there, and adds an invisible text layer positioned over the matching words. The page looks exactly the same, but you can now search, highlight and copy.
Getting accurate results
- Scan at 300 DPI. Lower resolutions blur small type and accuracy drops quickly.
- Keep pages straight. Rotate sideways pages with Rotate PDF before running OCR.
- Pick the document language. Recognition is tuned per language, and the wrong one produces garbled accents.
- High contrast helps: black text on white paper recognizes far better than faded print on colored stock.
A note on privacy
OCR PDF runs recognition in your browser, so your pages are not uploaded. The first time you use a language, the tool downloads that language's recognition data, which is a few megabytes. That download contains no part of your document.
Only need the words?
If you just want text from a photo or screenshot, Image to Text skips the PDF step and gives you plain text to copy. For a PDF that already has a text layer, PDF to Text extracts it directly without any recognition.