How to OCR a Scanned PDF on iPhone Without Uploading It

TL;DR

PDFExtract checks each page: real text layer pages are read directly, pages that are mostly images go through on-device OCR (Apple Vision). Raise the render resolution or switch to Accurate for small print; choose the languages first.

The problem

You have a scanned contract, a photographed textbook page or a PDF printed from a scanner. The PDF contains only pictures of text, so you can’t search, copy or quote it. Cloud OCR services read it, but only after you upload a document you may not want to share.

How PDFExtract decides

PDFExtract looks at each page separately:

  1. Is there a text layer that covers the page? If so, it uses that text exactly as written. It’s more accurate than any OCR.
  2. Otherwise it renders the page to an image and runs OCR on the device with Apple’s Vision framework.
  3. A page that carries only a tiny native text (a scanned page with a header watermark, say) counts as a scan, and OCR output replaces the native text only when it reads much more.

Mixed documents therefore work: digital pages stay perfect, scanned pages are recognised.

Steps

  1. In Settings, choose your Recognition Languages first (see the language guide).
  2. Import the PDF.
  3. Wait while it runs. Page-by-page failures don’t abort the run; the unreadable pages are listed afterwards.
  4. Review the Markdown. Copy or share it.

Getting better accuracy

  • Render Resolution: 200, 300 (default) or 400 DPI. Higher helps with small type and dense CJK, at the cost of time and memory.
  • Recognition Level: Accurate (default) or Fast.
  • Language correction on helps with ordinary prose; turn it off for codes, names and numbers.
  • Start with a clean scan: straight, high contrast, no shadows. Fix skew before importing.
  • If a run says it produced no readable text, rerun at a higher resolution.

Limits

OCR is a best guess. Check numbers, names and anything legal or financial against the original. Handwriting and very low-quality scans are unreliable. Free users get the first 5 pages.

FAQ

How does it decide between text and OCR? Per page. If the page’s embedded text covers enough of the page, it uses that exact text. Otherwise it renders the page and runs OCR.

Which OCR engine? Apple’s on-device Vision text recognition. Nothing is sent to a server.

What about a scan with a text watermark? A page with only a few native characters is treated as a scan, and OCR replaces them only when it reads clearly more text.

Why is OCR slow? Each page is rendered at 200–400 DPI and recognised on the phone. Long scans take minutes, so keep the screen on.

The app PDFExtract →