A scanned page starts as an image. Your PDF viewer can display the letters, but unless the file contains text data, search and copy have nothing to work with. Optical character recognition—OCR—analyzes the image and estimates the characters, words, and locations it sees.

The result is useful, but it can contain mistakes. OCR adds a text version of the page; it does not replace the original image or guarantee that every character is correct.

Three things people call a scan

TypeWhat it containsWhat you can do
Image-only PDFOne or more page imagesView, print, and annotate visually; text search usually fails.
Searchable PDFPage image plus an invisible text layerFind and often copy text while preserving the visual page.
Editable documentReconstructed text and layoutChange words and structure, but formatting may shift.

A searchable PDF is often the best choice for records: you still see the page image, and the hidden text lets you find things faster.

How OCR works

  1. Prepare the image. The scanner corrects perspective, rotation, contrast, and sometimes shadows so characters are easier to distinguish.
  2. Detect text regions. The recognition engine looks for lines, blocks, tables, and reading order instead of treating the page as one single picture.
  3. Recognize characters. It compares shapes against a language model and returns words with locations and confidence estimates.
  4. Place a text layer. The PDF keeps the image and stores recognized text in a position that lines up with it.
  5. Review uncertainty. Check by eye the words OCR is unsure about, especially numbers, names, dates, and legal language.

Why OCR errors happen

Recognition is sensitive to input quality. Small type, low contrast, motion blur, skew, shadows, decorative fonts, handwriting, stamps, checkboxes, and mixed languages all make mistakes more likely. A page you can read easily can still end up with a wrong character in the hidden text.

Searchable doesn’t mean correct. Use OCR to find and read text, then check the original page before you rely on a number or a name.

How to test whether a PDF is searchable

Open the exported file in a viewer that supports text search. Search for a distinctive word in the middle of the page, select a sentence, and paste it into a plain-text field. Then compare punctuation, numbers, and line order with the original scan. If you only test the title, you can miss errors in the main text.

  • Search finds a word that is not in the filename.
  • Selected text follows the visible page in a sensible order.
  • Decimal points, dates, and identifiers match the image.
  • The PDF still opens and prints correctly when the text layer is ignored.

How Paperlane approaches OCR

Paperlane runs OCR inside the app, as part of scanning, instead of treating it as a separate upload. Recognition is designed to run on device, then show you the words it’s unsure about so you can check them before you export a searchable PDF. You see which parts to check, instead of getting a block of text that looks right but hasn’t been checked.

Example

Imagine searching for a clause in a twenty-page scan and getting no result because every page is only an image. OCR adds a text layer behind the visible page so search and copy can work.

The text helps you search, but it doesn’t replace the scan. Test a heading, body sentence, name, date, and number, and review the visual page whenever the result affects a decision.

Questions people ask

Can I search an image-only PDF?

Not reliably. The viewer needs a text layer or another index. Running OCR adds searchable text while keeping the original page image.

Does OCR make a PDF editable?

No. Searchable text and editable layout are different outputs. OCR may let you copy text, but changing the original page structure generally requires a separate conversion tool.

Should I trust OCR for a contract or tax form?

Use it to find and review content, but verify critical names, dates, amounts, and clauses against the image before making a decision.

Google describes on-device text and document processing as useful for offline, privacy-sensitive mobile experiences in its ML Kit documentation.