A scanned PDF is a picture that happens to look like text — you can't search it, select a sentence, or copy a paragraph out of it, because as far as the file format is concerned, it's just an image. OCR PDF fixes that by reading the text out of each page and adding it back in as a real, searchable layer — entirely on your device.
This guide covers how the process works, what it does to your file, what to expect for accuracy, and how to get the best results from imperfect scans.
What OCR actually does to your file
OCR (optical character recognition) doesn't change how the page looks. It adds an invisible layer of real text positioned exactly behind the scanned image, so when you search, select, or copy in a PDF reader, you're interacting with that text layer — while what you see remains the original scan, pixel for pixel.
Under the hood, the process has two stages, and both matter for the result. First, the engine divides each page into regions and lines, working out where words sit — which is why a crooked page hurts before a single letter is even read: the boxes are drawn at the wrong angles, and characters bleed across boundaries. Second, it examines each word image and works out which characters it's looking at, choosing the most probable reading based on the word's overall shape and context. The recognized text is then placed at the same coordinates as the image it came from, which is why searching highlights exactly the right spot on the page rather than some approximate region.
How to run OCR on a PDF
- Open OCR PDF and drop in your scanned file.
- Let it process — this runs in your browser, so expect roughly a few seconds per page; a long scan can take a minute or more.
- Download the result. It looks identical to the original, but is now searchable and selectable.
The output is a normal PDF: same page count, same appearance, same print behavior — plus the text layer underneath. Any PDF reader from here on treats it as a text document for search and copy purposes.
Why "searchable" matters more than it sounds
The difference between an image-only scan and a searchable one sounds cosmetic until you hit it in practice. Consider what becomes possible:
- Finding things — Ctrl+F through a forty-page contract for "termination", a manual for a setting name, a medical record for a medication. Without a text layer, the only way is reading every page.
- Copying text out — quoting a clause into an email, pulling a figure into a spreadsheet, reusing a paragraph. With a scan, your only option is retyping it.
- Desktop search — file-search tools like Windows Search and Spotlight index text, not pictures. A folder of unsearchable scans is effectively invisible to them; once OCR'd, documents surface by their contents.
- Screen readers — for someone using assistive technology, an image-only PDF is a blank page. A text layer is what makes a scanned document accessible at all.
It's the difference between a document that stores information and one that lets you use it. The scan preserved the appearance of the text; OCR restores its function.
What affects accuracy
OCR does best on clear, high-resolution scans of printed text — a typical office scanner output at a reasonable DPI works well. Accuracy drops for handwriting, skewed or crooked pages, low-quality or blurry scans, and unusual fonts. Because this runs entirely on your device rather than a paid server-side engine, it won't match those services on genuinely difficult documents — but for ordinary printed scans, it's a real, useful upgrade from "just a picture."
Most accuracy problems trace back to the scan, not the software — which means most of them are fixable before you ever run the tool:
- Flatten and straighten the page — curled paper and a skewed feed angle are the top two accuracy killers.
- Scan at a reasonable resolution — too low and letters blur together; 300 DPI is the usual sweet spot for text.
- Prefer grayscale or black-and-white for text pages — color noise from a paper texture gives the engine extraneous detail to misread.
- Clean the glass and check for shadows — a smear or a finger-shadow across a line turns clear text into a guessing game.
And for documents where errors genuinely matter — legal, financial, medical — treat OCR output as a draft: read it against the page, at least for the critical figures. Recognition software confuses things a human never would (an "8" for a "B", a "1" for an "l"), and those slips concentrate exactly where proofreading is hardest: in numbers, which no spell-checker will flag.
English only, for now
This release supports English. If your document is in another language, the text layer may come out inaccurate or empty — more languages are planned for a future update.
When you don't need OCR
If your PDF already has selectable text — it was exported from Word, a website, or another digital source rather than scanned — you don't need this at all. Running OCR on a PDF that already has real text won't hurt anything, but it also won't add value.
The quick check: open the file and try to select a sentence. If it highlights, the text layer already exists. If your cursor drags a box or selects the whole page as one object, you're holding a scan — and this tool is what gives it one.
After OCR: what the text enables
Once a scan has a text layer, the rest of the toolkit works on it too. PDF to Word converts the recovered text into an editable .docx; copying quotes into anything else is a selection away. The scan keeps its original appearance throughout — OCR changes what the document knows, not what it looks like.