About OCR PDF
A scanned PDF looks like a document but acts like a picture. You can scroll through it, but you can't search inside it, can't copy a sentence, can't run a screen reader over it, and can't paste a paragraph into a translation tool. The text you see is just pixels arranged to look like letters.
OCR PDF reads those pixels with an image recognition engine and writes the recognised text into the PDF as a hidden layer behind the original image. The document looks identical to the source, but now it's searchable, selectable, and machine-readable. Same scan, suddenly useful.
What OCR does to your file
The visible page stays exactly the same. Same image, same look, same layout. Nothing about how the PDF appears in a reader changes.
Behind the image, the OCR engine places a layer of recognised text positioned to match the words in the picture. When you select text in the PDF, you're actually selecting that hidden layer. Same when you search; the search runs against the text layer.
This dual-layer approach is the standard for OCR-enhanced PDFs. It preserves the original look of the scan (which sometimes matters for legal or archival reasons) while making the content searchable.
Languages supported
The tool works out the language from the first pages, or you can pick it yourself. 78 languages are available: English, Hindi, Marathi, Sanskrit, Nepali, Bengali, Gujarati, Punjabi, Odia, Tamil, Telugu, Kannada, Malayalam and Urdu; Italian, Spanish, French, German, Portuguese, Latin, Russian, Polish and most other European languages; Chinese (simplified and traditional), Japanese, Korean, Arabic, Persian, Hebrew and more.
Pick the right language for the source document. The OCR engine performs noticeably better when it knows what alphabet and word patterns to expect. If the document mixes two languages (Hindi forms with English headings, for instance), pick the main one and add the other under “Also contains”.
When you'd OCR
Old contracts and historical documents that were scanned years ago and need to be searchable for an audit or a research project.
Receipts and invoices that arrive as scans. Once OCR'd, you can copy line items into a spreadsheet or feed them into accounting software.
Books, journals, and articles that you want to quote from. Selecting the paragraph and pasting it into a notes app is faster than retyping.
Government forms that came as scanned PDFs. Filling them in is easier when you can read the form fields with a screen reader or autofill tool.
Anything you want to translate. Google Translate, DeepL, and the like all need real text input, not images. OCR converts the image to text so the translator can do its job.
Accuracy and what affects it
Clean modern scans at 300 dpi or higher get accuracy in the high 90s percent range. The remaining errors are usually obvious to a reader and easy to fix manually if it matters.
Low-quality scans, especially from old fax machines or poor phone photos, drop the accuracy. The engine still does a useful job, but proofread the output if you're using the text for anything official.
Handwritten text is much harder. The engine can read printed text with high accuracy and clear handwriting with moderate accuracy, but messy handwriting or cursive often comes out garbled.
Skewed pages (where the scan was tilted) reduce accuracy. Straightening the page first, with the rotate tool, helps. Some scans benefit from a contrast-boost in an image editor before OCR.
Time it takes
OCR is the most compute-heavy of the tools here. A ten-page document takes about thirty to sixty seconds depending on page complexity. A hundred-page document might take five minutes or more. The progress bar shows where it is.
Running OCR in the browser is what keeps it private. Some cloud OCR services are faster but they need your scan to live on their server, which is a non-starter for sensitive documents.