Turn a scanned PDF into text
If a PDF-to-text tool gave you an empty file, the document is a scan: a stack of photographs with no characters in it. The only way forward is optical recognition.
OCR Scanned PDF — free and local
Recognises text in scanned PDFs — English, Chinese or both — with a confidence score.
Why the empty result is correct, not broken
A PDF produced by Word, LaTeX or Google Docs contains real characters plus font references. Extracting them is lossless. A scan contains only pixels, so there is nothing to extract — a tool that returns “plausible” text from a scan is doing OCR and not telling you.
This site keeps the two separate on purpose: PDF to Text reads the text layer and says “empty” when there is none; OCR guesses at pixels and reports a confidence score so you know how much to trust it.
Getting a usable recognition rate
Resolution matters most: 300 DPI scans recognise at 95–99% on printed text, while a 100 DPI fax drops sharply. If you control the scan, scan at 300 DPI greyscale rather than colour — smaller file, better recognition.
Language matters second: pick the language of the document, not your keyboard. For a Chinese document with English brand names, “Chinese + English” beats English alone, at roughly double the processing time.
Skew is the third killer. A page photographed at an angle recognises much worse than a flatbed scan; straighten the photo before converting it to PDF if you can.
What to do with the output
The result is per-page text with a confidence number, downloadable as a .txt. Paste it into Word if you need a document, or into a spreadsheet import if the scan was a table — though columns rarely survive OCR intact.
Check the low-confidence pages first. Anything below 70% usually has a systematic problem: wrong language, a rotated page, or a page that is mostly a photograph.
Questions
How accurate is browser OCR?
On a clean 300 DPI scan of printed text, typically 95–99%. On phone photos, handwriting, or faxes, much lower. Every page reports its own confidence.
Is my scan uploaded?
No. The recognition engine runs in your browser via WebAssembly. The engine and language model download once (~6–10 MB) and are then cached.
Can it produce a searchable PDF?
Not yet. An invisible text layer over a scan needs an embedded font for every script, which would add megabytes. This tool gives you the text itself.
Related
SlimPDF runs every tool in your browser. No file is ever uploaded, stored or shared.