How to make a scanned PDF searchable.

You have a PDF you cannot search. Ctrl+F finds nothing, you cannot select a sentence, and copying is impossible. The document looks perfectly normal on screen, which is what makes it confusing. Here is what is actually going on and how to fix it.

There are two kinds of PDF

A PDF made by a word processor, a browser, or almost any program carries an invisible text layer alongside the visual page. That layer is what lets you select, search, and copy. The characters are stored in the file as characters.

A PDF made by a scanner, a phone camera, or a photocopier is a photograph of paper wrapped in a PDF container. There is no text in it at all — only pixels arranged to look like text. Your eyes read it. Your computer sees a picture.

The test takes two seconds. Open the file and try to drag-select a sentence. If words highlight, you have a text layer. If nothing happens, or the whole page highlights as one block, you have a scan.

If text highlights, stop here

You do not need OCR, and using it would make things worse. The file already has exact characters stored in it; recognition would throw those away and replace them with guesses from an image. Use PDF to Text to pull the words out, or just search the document in your reader — it already works.

People sometimes reach for OCR because search is failing for an unrelated reason: the text layer exists but the reader has not indexed it, or the words are hyphenated across lines. Neither is fixed by recognition.

If nothing highlights, you need recognition

Optical character recognition looks at the picture and works out which shapes are letters. What you do with the result is where the important choice lies.

Plain text gives you a .txt file of the recognised words. Useful for quoting, pasting, or feeding into something else. You lose the layout entirely.

A searchable PDF keeps every pixel of the original scan and lays the recognised words invisibly on top, each one positioned over the printed word it matches. The document looks identical — same scan, same marks, same coffee stain — but Ctrl+F now works, text selects, and copying returns real characters. This is what people mean when they search for "OCR PDF to PDF", and for archiving or anything you need to find again it is almost always the right output.

The invisible layer is a documented PDF feature, not a trick. Text can be drawn in render mode 3, which means "lay this out but do not paint it." Every OCR product from Acrobat down does the same thing.

Doing it

Open OCR a PDF and drop the scan in. Three steps, matching the tool:

1. Drop a scanned PDF — the kind where you cannot select any text.
2. Let it recognise each page — pages are rendered and read one at a time, on your device.
3. Download — either the searchable PDF, which looks identical to the original, or the plain text as .txt.

There is a limit worth knowing before you start: 200 pages. Every rendered page is held in memory until the file is saved, and beyond roughly that point the browser cannot complete the save. For a longer scan, split it with Split PDF, run each part, then merge the results. The tool refuses up front rather than working for ten minutes and failing at the end.

What accuracy to expect

A flat 300 DPI scan of ordinary printed text recognises at 99 percent or better. A phone photograph taken square-on in good light is close behind. Below that it degrades, and it degrades in a way that is easy to miss: the output still looks like text.

The reliable failures are skew, low resolution, and shadow. A page photographed at an angle has converging lines; below about 20 pixels per line of text the letterforms stop being distinguishable; a shadow across half the page destroys the contrast the engine relies on. Handwriting does not work — the engine is trained on printed type, and neat block capitals are the most you can hope for. There is more detail in what OCR can and cannot do.

Numbers deserve particular suspicion. A misread letter produces a nonsense word you notice. A misread digit produces a plausible figure you do not.

Why it is slow

Recognition runs in your browser, using Tesseract compiled to WebAssembly. The engine and its language data are about 6 MB and download once, then stay cached. After that, each page is rendered and read by your own processor — a couple of seconds on a laptop, noticeably longer on a phone.

A server would be faster. It would also mean sending the document somewhere, and scans are disproportionately the documents you would least want to send: medical letters, contracts, bank correspondence, identity papers. The whole reason a scan needs OCR is that the words are locked in a picture, and that need does not care how sensitive the picture is.

It is not uploaded

Nothing about the document leaves the tab. You can confirm that rather than trust it: open developer tools, switch to the Network panel, and run the recognition. You will see the OCR engine download once, coming in. Nothing goes out. The general method works on any OCR site, including the one you use now.

What to do next

If the paper is still in your hand, do not scan it with a scanner app first. Scan to PDF photographs the pages with your camera and builds the document in the tab, and then you run this on the result.

If the file is too big to send, be careful. Compressing a PDF re-renders the pages as images — which destroys the text layer you just spent ten minutes adding. If you need both a small file and searchable text, compress first and OCR second, never the other way round.

Common questions

Tools on this page

• OCR PDF — Make a scanned PDF searchable, or pull the text out.
• Scan to PDF — Photograph pages with your camera, get one PDF.
• PDF to Text — Pull the words out as plain .txt.