How to copy text from a scanned PDF

Short answer: A scan has no text, only a picture, so you first need OCR (optical character recognition). Open the file in PDF OCR, choose the document's languages, run it and copy or save the text. A clean, straight scan of printed text is recognised much better than a tilted, blurred or noisy one, but accuracy also depends on the language, the font and the content: on our one short test page the clean English scan had no errors, while the clean Chinese scan had 7.8% errors when punctuation is counted. Always proofread numbers and names.

Last updated:

Steps

  1. Open the scan in PDF OCROpen the PDF OCR tool on a desktop browser and choose the scanned PDF, or a JPG or PNG photo of a page. The file stays in your browser.
  2. Choose the document languagesTick the languages the document is written in (English, Simplified or Traditional Chinese, Japanese, Thai) — not the language of the website. Fewer languages give better results.
  3. Choose the pages and runEnter the pages to recognise (up to 100 per run) and select Recognise text. Each page takes a few seconds.
  4. Copy or save, then proofreadCopy the text from the box or save it as a TXT file. Compare numbers, names and amounts with the scan before you use them.

Why you cannot just copy it

A scanner or phone camera produces a picture of the page. The PDF holds that picture and nothing else, so there is nothing to select or copy. You can confirm this with the text-layer check: it reports “No text found (page has images)”. OCR looks at the picture, recognises the letters and gives you text.

How good is it? Our measurements

We printed the same six-line delivery note into scans of different quality (scan-quality-test.pdf, Chinese version) and ran our PDF OCR tool on them in Chromium on 12 November 2026. Character error rate = wrong, missing or extra characters divided by the characters in the original (spaces not counted):

PageEnglish: errorsEnglish: lines exactly rightChinese: errors (letters and digits only)
300 dpi, clean0%6 of 61.1%
150 dpi, clean0%6 of 61.1%
100 dpi, clean0%6 of 61.1%
150 dpi, tilted 1.2°, blurred, speckled, heavy JPEG85%0 of 681%

What this shows, and what it does not:

  • Resolution was not the problem for clear, printed text: even 100 dpi came out right. Resolution matters more for small print.
  • Damage was the problem: a slightly tilted, blurred, speckled and heavily compressed page produced unusable text in both languages. Improving the scan beats any setting.
  • Chinese punctuation is often recognised as its half-width form (: → :) and the dash — as the character 一; that one character was the 1.1%. Counting punctuation as well (the same rule as the English column), the clean Chinese pages had 7.8% errors and the damaged one 83%.
  • This is one test page in a clean sans-serif font — a sample, not a benchmark. Your documents can do better or worse.

Make the scan easier to read

  • Scan at 300 dpi in greyscale if you can; it costs little and helps with small print.
  • Keep the page flat and straight; photograph it from directly above in even light.
  • Avoid heavy JPEG compression (“small file” settings).
  • Choose only the languages that are really in the document.
  • Thai: tone marks above upper vowels are often missed (ที่ is read as ที); proofread Thai text closely.

Other ways

WayUpload?Notes
PDF OCR on this siteNoDesktop browsers; English, Chinese, Japanese, Thai; up to 100 pages per run
Make PDF searchable on this siteNoKeeps the PDF, adds an invisible text layer
Adobe Acrobat Pro: Scan & OCR › Recognize textNo (desktop app)Paid; also edits the recognised text
Online converters and cloud drivesYesConvenient, but the file goes to someone else’s server — see is it safe to upload PDFs to online tools

Whatever you use: OCR output is a draft. Proofread it against the scan before you rely on it.

FAQ

Do I want the text, or a searchable PDF?

If you need the words elsewhere (an email, a document), use PDF OCR and copy the text. If you want to keep the PDF but be able to search and select in it, use Make PDF searchable: it adds an invisible text layer and the pages look the same.

Why does it not work on my phone?

Recognition loads 1–3 MB of language data per language and needs a lot of memory per page, so the OCR tools run on desktop browsers only for now.

Can OCR read handwriting?

Not reliably. The recognition models used here are trained for printed text. Handwriting usually comes out wrong or not at all.

Is my document sent to a server for recognition?

No. The recognition engine (Tesseract) and the language data are downloaded from this site into your browser, and the page images are recognised there. We recorded every request the page makes; see is it safe to upload PDFs to online tools.

Desktop app for Windows

The Windows installer has not been published yet. There is no download link on this site until it is.