How to OCR a PDF
- Drop your scanned PDF. Photographs of pages work too, though a flat scan reads better.
- Set how many pages to read. Recognition is slow, so start small if the document is long.
- Run OCR, then download. You get the recognised text on screen and a searchable copy of the file.
What a scan is missing
A scanned page is a photograph. To a computer it contains no words at all, so searching it finds nothing, you cannot select a line to quote it, and a word counter reports zero. OCR looks at the shapes, works out which letters they are, and writes that back into the file as text you cannot see but every viewer can find.
Because the text layer is invisible and sits precisely over the original pixels, nothing about the document's appearance changes. Search for a word and your reader will highlight it in the right place on the scan. The same text layer is what makes a scan readable to screen readers, which a photograph of a page otherwise isn't at all.
Why this one is slower than the upload sites
Server-based OCR is fast because it runs on hardware built for it. This runs on whatever device you happen to be holding, which is slower and means your document is never handed to anyone. For a contract, a medical record or an old family document, that is usually the trade worth making. For a thousand-page archive it is not, and desktop software is the better answer.
Once a document is searchable, extract the text to work with it elsewhere, or count the words, both of which return nothing on a scan until OCR has run.
Common uses for OCR
- Search an old scan: find a clause in a contract that was scanned years ago.
- Quote from a scanned book: select and copy text that was previously just pixels.
- Make an archive findable: turn a folder of scanned records into documents you can search.
- Count words in a scan: get a word count from a document that reports zero until OCR runs.
Frequently asked questions
How do I make a scanned PDF searchable?
Drop the scan, choose how many pages to read, and click Run OCR. Each page is recognised and the words are written back into the file as an invisible layer, so the document looks identical but can be searched, selected and copied.
Does OCR change how my document looks?
Not at all. The recognised words are drawn at zero opacity, positioned over the pixels they came from. Your scan is untouched underneath; the text is now there to be found.
Why is it so slow?
Because a full recognition engine is running on your own processor rather than a server farm. Expect a few seconds per page, longer on a phone. That's the price of the file never leaving your device, and it's why the page count is capped by default.
Does the recogniser come from somewhere else?
No, and this took deliberate work. The usual browser OCR library fetches its language data from a public CDN, which would mean a third-party request on a site that promises none. Everything is served from this domain instead, so the first run downloads about 6 MB from us and then works offline.
How accurate is it?
Good on clean, straight, reasonably high-resolution scans of ordinary printed text. Poor on handwriting, heavy skew, low resolution, or photographs taken at an angle in bad light. Nothing corrects for a bad source, so a better scan is worth more than better software.
Which languages does it read?
English at the moment. The language data is a separate file per language, and each one is a few megabytes, so languages will be added one at a time rather than all at once.