Scanned PDF to Text
You opened a PDF. You tried to copy a sentence. Nothing happened. That's a scanned PDF: the pages are images, so there is no text to copy. OCR reads the pixels and writes the words back out. Drop the file below and you'll get a .txt you can edit, search, or paste anywhere. The file stays on your machine the whole time.
Drop your scanned PDF here
or click to browse
Accepts: .pdf
How do you end up with a PDF you can't search?
Almost always: a scanner. A printer-scanner combo, a phone app like Adobe Scan or Microsoft Lens, an office multifunction. They take a picture of each page and wrap the pictures in a PDF wrapper. The file looks like a normal PDF. It opens, it prints, it emails. It just doesn't have any actual text inside, only pixels.
You'll also hit this with faxes saved as PDF, scans from court records or library archives, and screenshots of pages someone glued together with a "convert to PDF" tool.
What OCR is actually doing
It binarizes the page (black ink, white paper), straightens any tilt, finds the lines of text, splits them into words, then runs each character region through a model that knows what letters look like. The output is plain text in roughly the order a person would read.
The engine here is Tesseract, the open-source OCR project Google funded for a decade. We use the WebAssembly build (Tesseract.js), which is why it can run in a tab instead of a server.
Nothing is uploaded
Most "free PDF OCR" sites take your file, send it to their backend, run OCR there, and email you a link. Fine for a flyer. Not great for a tax return, a medical record, or a contract.
Want proof? Open DevTools, switch to the Network tab, then hit Extract Text. You'll see two downloads on the first run: the OCR engine (around 10 MB) and the English language data (around 4 MB). Both are cached after that. Your PDF itself never goes anywhere.
Getting better results
- Resolution is the whole game. A 300 DPI scan reads cleanly. A 150 DPI scan produces noticeably more typos. If you're scanning fresh, 300+ DPI.
- If you can't re-scan, try the 3x render scale. It gives OCR more pixels to chew on.
- Phone photos of book pages: lay the book flat, kill the shadows, hold the camera square to the page. Curved spines and hand shadows are the biggest accuracy killers.
- Handwriting won't work. Tesseract is trained on printed text.
- Two-column layouts (academic papers, magazines) sometimes get read straight across instead of column by column. If you see jumbled output, crop the PDF to one column and run it twice.
Want a searchable PDF instead of a .txt?
Same engine, different output: Make PDF Searchable keeps the original page images and tucks the OCR text underneath as an invisible layer. Highlighting and Cmd-F start working, and the file still looks like the scan you started with.
A few things people ask
What about non-English documents?
This page only loads English. Mixed-language pages will get the English right and turn other scripts into garbage. More languages may show up later.
Is there a page limit?
No fixed limit. On a laptop, several hundred pages is fine. On a phone, things get tight past about 100 pages because of memory.
Does it keep formatting?
No. Bold, italic, columns, fonts: gone. You get raw text with line breaks in reading order. If you want the original look kept, use the searchable PDF tool.
Why is the first page slow?
That's the engine and language data downloading. After the first page, the rest run at the normal speed. Future PDFs reuse the cache.