OCR PDF — Make a Scanned PDF Searchable

RUNS ON THIS DEVICE

Run text recognition on a scanned PDF and download a copy you can search and copy from — the scan looks identical, but the words underneath are real.

The problem with scans: you can't Ctrl+F them

A scanned contract, an old report, a signed agreement that came back by scanner — each page is just a photograph, so nothing is searchable, nothing is selectable, and document systems can't index a word of it. OCR fixes that: the text in the images is recognized and written into the file as an invisible layer, positioned exactly over the printed words. The document looks pixel-for-pixel the same, but Ctrl+F works, text can be copied out, and search tools finally see the content. Elsewhere this is routinely a paid, premium feature; here it runs free on your own machine.

How the recognition works

Each page is rendered at high resolution and read by Tesseract, the open-source OCR engine, compiled to WebAssembly and self-hosted by this site — after the one-time engine download (about 15 MB), even that needs no further network access. Recognition takes a few seconds per page. The recognized words are then placed over the original pages — which are copied untouched, preserving scan quality — at the exact position of the printed text, so selecting a sentence in the output highlights the right place on the page. Seven languages are supported; picking the right one matters a lot for accuracy.

Private by architecture, honest about limits

Scanned documents are the most sensitive files people hold — signed contracts, medical letters, ID copies — and this tool never transmits them: rendering, recognition, and rebuilding all happen in your browser. Two limits stated plainly: OCR accuracy depends on scan quality (a sharp 300 DPI scan reads nearly perfectly, a blurry phone photo doesn't), and the invisible layer uses the PDF standard's built-in fonts, which cover Latin scripts — for Hindi or Arabic documents, recognized words outside that character set are counted and reported rather than silently dropped.

Frequently asked questions

Is my scanned document uploaded for OCR?

No. The OCR engine runs in your browser via WebAssembly, self-hosted by this site. After the one-time engine download, the recognition itself makes no network requests — your scan never leaves the device.

Will the PDF look different afterwards?

No. The original pages are kept untouched — the recognized text is added as an invisible layer on top, so the file looks identical but becomes searchable and selectable.

How accurate is the recognition?

On a clean, straight scan at reasonable resolution, very accurate. Blurry photos, handwriting, and skewed pages reduce accuracy — OCR reads print, and garbage in produces garbage out.

Which languages are supported?

English, Spanish, French, German, Portuguese, Hindi, and Arabic. Choose the document's language before starting — it significantly changes accuracy. The searchable layer itself is limited to Latin characters, and any words outside that set are reported.

Why is there a 50-page limit?

Recognition takes a few seconds of real computation per page, all on your device. Fifty pages keeps the run manageable; split longer documents, OCR the parts, and merge them back — every step stays local.