How to OCR a PDF (Turn Scans into Text)

Is OCR the right tool?

OCR recognizes text in pages that have no readable text layer — scans, photographs, and faxes. The test is simple: if you cannot select or copy the words on the page, the text is trapped inside images and OCR is the right tool.

If you can already select the words, the PDF has a text layer and PDF to Text will convert it exactly and instantly. OCR of a text-based PDF would slow things down and introduce recognition errors for no benefit — the tool even notices and suggests the faster route.

What OCR can and cannot do

OCR turns each page image into text via Tesseract. Printed documents at a decent scan resolution (150–300 DPI) usually recognize almost everything. Blurred pages, unusual fonts, handwriting, and low-resolution scans will have errors, so always skim the result before relying on it.

The output is a plain-text result you copy or download as .txt. Your original PDF is never modified, and the tool does not embed a text layer back into it. If you need the text inside a new PDF, paste the recognized text into Text to PDF, or open it in a word processor and export a fresh PDF.

Choose the language

Pick the document's main language from the 12 supported: English, Spanish, French, German, Italian, Portuguese, Russian, Hindi, Arabic, Chinese (simplified), Japanese and Korean. The first run of a language downloads that language's recognition model (~1.5–3 MB) into your browser from this site and caches it, so later runs are instant.

OCR is fully client-side: your PDF is rendered and recognized in your browser and never uploaded. Files up to 100 MB and 200 pages are supported; the model download on first use needs an internet connection.

After OCR

Skim the per-page result, fix obvious misreads, then copy or download. Because each page is recognized independently, a page marker (--- Page N ---) keeps the reading order clear in the .txt. Multi-column pages may read across both columns in each line; if the layout is complex, editing order in the text is usually faster than re-rendering.