A scanned contract, an old report photographed with a phone, or a stack of invoices from a copier: they look like documents, but to the computer each page is just a picture. You can't search them, select a sentence or copy a figure. OCR PDF recognizes the text on each page and adds it as an invisible layer, so the file looks exactly the same but becomes searchable.
OCR (optical character recognition) works on the image of each page, so its accuracy depends on the scan. It doesn't rebuild the layout into an editable document; for that, run the result through PDF to Word afterward.
Step-by-step
- Open the OCR PDF tool, or click the "OCR PDF" card on the home page.
- Click the drop zone or drag in your scanned PDF.
- Choose the "Document language". It starts with the language of the site you're using.
- Click "Process now." The first time you use a language, the browser downloads a recognition model of 2 to 15 MB and keeps it for next time.
- The status line shows which page is being recognized. Each page takes a few seconds, depending on your device. Pages that already have selectable text are left as they are.
- Your browser downloads the PDF with "-ocr" added to its name. Open it and search for a word you can see on a page to check the result.
Common mistakes
- Expecting perfect text from a poor scan. Blurry, tilted or low-contrast pages produce errors. Rescan at 300 dpi if the result matters.
- Choosing the wrong language. The model for one language misreads letters and accents of another. Pick the language the document is written in.
- Processing a password-protected PDF. The tool can't read it. Remove the password with Unlock PDF first.
- Expecting an editable document. OCR adds searchable text but keeps the page as an image. Use PDF to Word on the result to edit the text.
- Running OCR on handwriting. Tesseract is trained on printed text; handwritten notes are mostly not recognized.
FAQ
Is my file uploaded to recognize the text?
No. Recognition runs in your browser with tesseract.js, the Tesseract OCR engine compiled to WebAssembly. The only downloads are the program itself and a language model, never your document.
Which languages can it recognize?
English, Spanish, Portuguese, French, German, Indonesian, Simplified Chinese, Japanese and Arabic. Choose the document's language before processing; for Chinese, Japanese and Arabic, English is recognized too, so mixed words come out right.
Can I edit the text after OCR?
The PDF keeps its original appearance, with the recognized text as an invisible layer you can select, search and copy. To edit the text, run the result through PDF to Word, which extracts that layer into a .docx file.
Why are some words wrong?
OCR accuracy depends on the scan. Blurry, tilted or low-resolution pages, handwriting and decorative fonts produce more mistakes. For best results, scan at 300 dpi, keep the page straight and use good contrast.
Checklist
- The scan is sharp, straight and at least 200 to 300 dpi
- The document's language was selected
- The PDF has no password
- A word from the page was searched in the result
- Figures and names that matter were checked by eye