Optimize
OCR PDF
Add a searchable text layer to scanned documents.
Processed locally — your files never leave your device
How to use OCR PDF
- 1
Open your scanned PDF.
- 2
Pick the document's language and a recognition resolution.
- 3
Choose the output: a searchable PDF that looks identical, or plain text.
- 4
Click Run OCR — recognition happens on your device.
About this tool
A scanned PDF is a photograph of words: you can read it, but your computer can't. Search finds nothing, copy-paste grabs nothing, screen readers fall silent. OCR (optical character recognition) fixes that by analyzing the page images and recognizing the characters in them — and this tool does it entirely in your browser via WebAssembly, with no document upload, using the Tesseract engine that powers countless production OCR systems.
The searchable-PDF output is the clever part: your pages remain visually identical — the same scan, untouched — while an invisible text layer is placed precisely over the printed words. The result looks like the original but behaves like a real document: Ctrl+F finds things, text selects and copies, and search indexers can read it. Choose plain-text output instead when you just want the words out.
Accuracy depends on the scan. Clean, straight, 200+ DPI scans of printed text recognize at well over 95%; phone photos at odd angles, faxes and unusual fonts do worse. Eight languages are available, and picking the right one matters — recognition models are language-specific. The first run downloads the engine and language data (~15 MB); after that it's cached. Recognition is compute-heavy, so a long document takes a few minutes — the page counter keeps you posted.
Frequently asked questions
Will OCR change how my document looks?
Not at all in searchable-PDF mode. The recognized text is added as an invisible layer aligned over the scan — pages stay pixel-identical while becoming searchable and copyable.
How accurate is it?
On clean printed scans, typically 95–99% at word level. Accuracy drops with skewed phone photos, low resolution, handwriting (not supported) and decorative fonts. Raising the recognition DPI helps marginal scans.
Why is my document being processed so slowly?
OCR is genuinely heavy computation, and it's running on your device rather than a server farm — that's the privacy trade. Expect a few seconds per page, longer at high DPI.
What does the ~15 MB download mentioned in the tool include?
The Tesseract WebAssembly engine and your chosen language's recognition model, fetched once from a CDN and cached by your browser. Your document itself never goes anywhere.
Which languages are supported?
English, Spanish, French, German, Portuguese, Italian, Dutch and Hindi. Pick the document's main language — using the wrong model significantly hurts accuracy.