OCR vs. extracting text from a PDF: they're not the same tool

"I need to get the text out of this PDF" sounds like one task, but it's actually two, and using the wrong one either wastes time or silently fails. The difference comes down to how the PDF was created in the first place, and it's worth 10 seconds to check before picking a tool.

If the PDF came from Word, Google Docs, a webpage printed to PDF, or almost anything generated digitally, the text inside it already exists as text — actual characters with positions on the page, not pixels. Open it and try to click-and-drag to select a sentence: if a cursor highlights individual words, that text layer is already there, sitting in the file, waiting to be read back out. Pulling it out is just parsing — no recognition, no guessing, no errors, and it happens in a fraction of a second per page. That's what our Extract Text from PDF tool does: it reads the text layer pdf.js already knows how to parse and hands it back as plain text.

If instead the PDF is a scanned paper document — someone ran it through a scanner or photographed it with a phone — there is no text layer. The whole page is one image, and what looks like text is just a pattern of pixels. Try to select a sentence and nothing highlights, because there's nothing there to select. Getting text out of that requires actual optical character recognition: rendering the page, then running a recognition model that identifies letter shapes and guesses what word they form. That's what OCR PDF does here, and it's fundamentally a different, slower, and imperfect process — accuracy depends heavily on scan quality, and handwriting or a skewed low-resolution photo will produce real mistakes a text-layer extraction never could.

Running OCR on a PDF that already has a text layer works, technically, but it's strictly worse than extraction: slower, and it introduces recognition errors into text that was already perfectly accurate. Running text extraction on a scanned image, on the other hand, returns nothing at all, because there's no text layer to read — which is the one signal worth checking for before you pick a tool, since it tells you immediately which category your file is in.

Both run entirely in your browser either way, which matters more for OCR than extraction: a scanned document is disproportionately likely to be the kind of file people hesitate to upload anywhere — filled-out forms, IDs, signed pages, medical intake sheets. The category of document that most needs OCR is also the category where a browser-based tool being the only copy that ever touches the file matters most.