What this does

A scanned PDF is a picture of a document. It looks like text, but there is no text in it — searching finds nothing, selecting selects nothing, and copying copies nothing. This reads the picture, works out what it says, and writes those words back into the file as an invisible layer sitting exactly where they appear on the page.

Nothing about the page changes. The scan is still the scan, at the resolution it always was — the original pages are copied through untouched rather than redrawn, so nothing is re-compressed or flattened. What changes is that the document is now searchable, and you can select a paragraph and copy it.

Nothing is uploaded — and that matters more here

The reading is done by a WebAssembly build of Tesseract running inside this page, on your device. Your document is never sent anywhere, and there is no server on the other end to send it to. Every other OCR service on the web works the opposite way round: you hand them the file, their machines read it, and they hand it back. The documents people OCR are scans of passports, contracts, medical letters and bank statements, which makes "we processed it on our servers" a considerably bigger deal than it sounds.

The one thing that does come off the network is the reader itself — about 7 MB of engine and English language data, downloaded from this site the first time you use this page and cached by your browser afterwards. It comes from here rather than a third-party CDN, so no one else learns you opened this page either. Once it has loaded you can disconnect entirely and the tool carries on working.

How to use it

Drop in a scanned PDF and press the button. Progress is shown page by page; you can stop part way and nothing is written.
Normal or careful. Normal reads each page at 200 dots per inch, which is where accuracy stops improving much on an ordinary document. Careful reads at 300, which is slower and noticeably better on small print, faint photocopies and phone snaps of paper.
Pages that already have text are skipped unless you say otherwise. Reading them again would lay a second, worse copy of their own words over the top, and then every search for a phrase would find it twice. A mixed document — a typed report with scanned annexes — gets the treatment page by page.
You can take the plain text instead. The .txt holds what was read, page by page, in reading order.

What it can and can't read

It reads printed English well — books, letters, reports, forms, receipts — including pages that were scanned sideways or upside down, since the layer is written along whatever direction the words actually run in. It is much weaker on handwriting, on heavily stylised type, on very low-resolution scans, and on pages where the text sits over a photograph.

Accented characters common in Western European languages come through as they are. Characters outside that range — Polish, Czech, Turkish, Greek, Cyrillic, anything non-Latin — lose their marks or are dropped, because the invisible layer is written in a standard PDF font that cannot encode them. Searching for the plain spelling still finds the page.

Words the engine isn't reasonably sure about are left out rather than guessed at. A wrong word in an invisible layer is worse than a missing one: you can't see it, so you never correct it, and it makes the document findable under something it doesn't say.

One thing it won't do for you is straighten a page. A scan that displays on its side or upside down reads badly, because the engine is looking at sideways letters — turn it the right way up in Organize PDF first, then bring it back here. A page that displays upright is fine however it is stored: the layer is written along the page's own axes, so a rotated page gets rotated text and searching works either way.

Frequently asked questions

Why is the first page slow?

Because the reader is being downloaded and started up. After that each page takes roughly a second on a normal document, or two to three on careful.

Does the file get bigger?

A little. The text layer is only text, so it adds a few kilobytes per page — nothing next to the scanned images, which are untouched.

Can I search the result in any reader?

Yes. The layer uses the standard invisible text rendering mode, which is what every OCR program produces and what every PDF reader, search index and pdftotext expects.

My PDF already has text and nothing happened.

Then it didn't need reading — it's already searchable. Tick "Read pages that already have text" if you want it read anyway, though usually the text already in a PDF is more accurate than anything OCR would produce from a picture of it.

It says my PDF is protected.

Encrypted PDFs can't be modified without the password, and that includes files with an owner password that only restricts printing or copying. Open it in a reader that can unlock it, save an unprotected copy, and use that.