OCR a Scanned PDF in Your Browser: How It Works and Its Limits

When a document is created digitally — exported from a word processor, say — it carries a text layer: the actual characters, which is why you can highlight, search, and copy them. A scan or a photo of a page has none of that. It is an image of text, and to the file there is no difference between a paragraph and a picture of a cat; both are just arrangements of coloured dots.

That is why searching a scanned PDF turns up nothing and why copying “text” from it is impossible. Nothing is broken — there simply are no characters in the file to find. OCR is what creates them.

OCR examines the image, detects where the text is, and recognises the shapes of the letters and words, producing machine-readable characters from the picture. The tool here then attaches those characters back onto the page as an invisible text layer sitting exactly over the visible words — so the document still looks identical, but is now searchable and selectable — or hands you the recognised text on its own if that is what you need.

All of this happens in your browser. The recognition engine and its language models are downloaded to your device from this site, and the actual analysis runs locally on your machine; your scanned document is never uploaded. That is unusual for OCR, which has traditionally been a server-side, send-us-your-file affair — here it stays on your device, which matters a great deal when the thing you are digitising is a contract or a medical record.

The tool recognises English and the common Latin-alphabet languages — French, German, Spanish, Portuguese, Italian, and Dutch among them — and also handles Simplified and Traditional Chinese and Japanese, so mixed Latin-and-CJK documents are covered. You pick the language of your document before running it, which helps the engine choose the right model and improves accuracy.

It is worth being upfront about one gap: Korean is not currently supported, because the underlying engine version does not include a Korean model. If your document is in Korean, this particular tool is not the right fit. For the languages it does cover, selecting the correct one before you start makes a noticeable difference to the result.

On a clean, straight, reasonably high-resolution scan of printed text, modern OCR is very good — good enough that most words come through correctly and the document becomes genuinely searchable. Pages that are slightly rotated are handled, since the tool detects and corrects for orientation. That covers the common case of a decent office scan.

Accuracy drops where the input fights back. Low-resolution or blurry scans, heavy background patterns or coloured paper, very small type, unusual decorative fonts, and especially handwriting all push error rates up, because the letter shapes become harder to tell apart. OCR is a best-effort reading, not a guaranteed transcription, so for anything important — names, figures, legal wording — treat the output as a strong draft and proofread it against the original rather than trusting it blindly.

Once a document has been through OCR you have options. Keeping the searchable-PDF form is ideal when you want the page to look exactly as scanned but also be findable and copyable. Taking the plain text instead is better when you only need the words — to paste into an email, quote, or feed into another tool. And because OCR gives a previously image-only file a real text layer, it also unlocks the next steps: an OCR’d PDF can then be extracted as clean text or converted toward an editable Word document, neither of which was possible while it was still just a picture.