loading tool…
What OCR PDF does
OCR PDF reads the text in a scanned PDF with a recognition engine running in your browser and lays it invisibly over each scanned page, so the scan becomes selectable and searchable while looking exactly the same.
A scanned PDF is a stack of photographs. The words are visible but not real — you can't select them, search them, or copy them, because as far as the file is concerned there is no text, only pixels. OCR (optical character recognition) reads those pixels and turns them back into actual text.
Nearly every free OCR site does this by uploading your scan to a server, which is exactly where a confidential document shouldn't go. This one runs the recognition engine — the same open-source tesseract engine those servers often run — in your browser through WebAssembly. The engine itself is served from this site, not a CDN, and your file never leaves the page. The recognized words are drawn invisibly over each page of the original, so the document looks untouched but reads as text.
How to use OCR PDF
Upload the scanned PDF
Drop in the scan. Password-protected files are refused — unlock first if needed.
Run recognition
Click Make PDF searchable. The first run downloads the ~7 MB engine; recognition then takes a few seconds per page, with progress shown.
Check the result
The searchable PDF downloads automatically. The tool reports the engine's own confidence score, how many words it added, and shows what the first recognized page looks like to the engine — read that preview before trusting the rest.
OCR PDF: details and limits
Accuracy tracks scan quality, not wishful thinking: clean, straight, high-resolution printed text recognizes well; faxes, phone photos, small print, columns and handwriting recognize badly. There is no fixed accuracy number that would be honest to print here — the tool reports the engine's own confidence for your specific document instead.
Dubious words are dropped rather than buried invisibly in your file, and the tool tells you how many: anything the engine scored below 30% confidence, plus the rare word the built-in font can't write. Low-confidence output is mostly noise that would pollute search results.
English (Latin script) only for now. The invisible layer is written with the same built-in font as the site's other tools, so non-Latin characters could not be stored even if the model read them.
Pages that already have selectable text are skipped and reported — OCRing them again would put a second, slightly different copy of every word under your selection cursor.
The original pages are never re-encoded or replaced. The scan image stays byte-for-byte what it was; the only change is invisible text drawn on top. Selection boxes can sit slightly off the printed word — that's the recognition's coordinates, not a change to your page.