What Is OCR, and How Does It Turn a Scanned PDF Into Text?

If you've ever tried to select or copy text from a scanned document and found that nothing happens — no highlighting, no cursor, nothing — you've run into the difference between a picture of text and actual text. Optical character recognition, or OCR, is the technology that bridges that gap.

Why a scanned PDF doesn't have real text

When you scan a paper document or photograph a page, what you get is an image — a grid of pixels that happens to look like text to a human eye, but means nothing to a computer beyond colors and shapes. A PDF built from scanned pages is really just a container for those images, one per page. There's no underlying character data for a computer to select, search, or read aloud, even though a person looking at it can clearly read every word.

What OCR actually does

OCR software analyzes the shapes on the page and matches them against patterns it has learned for each letter and number, effectively “reading” the image the way a person would, then outputting the result as real, computer-readable text. Modern OCR engines are trained on enormous datasets of text in different fonts, sizes, and conditions, which is why they can handle everything from a crisp printed page to a slightly skewed, mediocre-quality scan.

The tool on this site uses Tesseract.js, an open-source OCR engine originally developed at HP and later maintained by Google, compiled to run inside a web browser via WebAssembly. That last part matters: it means the actual text recognition happens using your own device's processor, not a remote server, so a scanned document's contents are analyzed locally rather than being sent anywhere to be read.

What OCR is good at — and where it struggles

• Good: clean, printed text — typed reports, forms, printed contracts, book pages — especially when the scan is reasonably high-resolution and not skewed.
• Good: making an archive of scanned documents searchable, so you can find a specific document by its contents instead of opening each one manually.
• Struggles: handwriting, which most general-purpose OCR models (including the one used here) aren't trained to recognize reliably.
• Struggles: low-resolution, blurry, or heavily skewed scans, where character shapes become ambiguous.
• Struggles: unusual fonts, dense tables, or multi-column layouts, where the reading order can get scrambled.

A typical OCR workflow

In practice, extracting text from a scanned PDF looks like this: upload the file, let the OCR engine process each page (this takes longer than most PDF operations, since it involves genuinely analyzing every character), then review and copy the resulting text. It's worth skimming the output for recognition errors, especially around numbers, punctuation, and any unusual formatting, before relying on it for something important.

OCR won't perfectly digitize every document, but for the common case — a clearly printed page that just happens to be trapped inside an image — it turns an unsearchable scan into text you can copy, search, and edit in seconds.