What Is OCR and When Do You Need It? A 20-Page Scan, Measured

OCR in plain words

A scanned page is a photo of paper. You can read the words, but to a computer they are gray dots, like any other picture. OCR, short for optical character recognition, finds the shapes of letters in that picture and turns them into real text.

In a PDF, that text goes into an invisible layer behind the picture, each word laid over its own image. The page itself stays a picture. Now, though, a word can be selected and copied, Ctrl+F finds it, and a screen reader has something to read.

Without that layer, a scan is closed to assistive technology. The W3C writes the web accessibility guidelines, and its technique for scanned PDFs says: “Assistive technologies cannot read or extract the words.” When the original file is gone, the fix it names is OCR.

Three checks: does your PDF need it?

Open the PDF in any viewer and try to select one word. If the words highlight one by one, the file already has text. If the drag draws a box over the whole page, or nothing happens, the page is a picture.

Then search for a word you can see on the page, with Ctrl+F, or Cmd+F on a Mac. No match for a word in plain sight means there is no text behind it. If a copied line pastes as gibberish, the layer is usually there but wrong.

OCR PDF runs the same check for you. Under the preview, the table “Check of the finished file” shows “Text layer, p. 1” before and after, for the page you have open. On our scan it read “no” before and “yes” after; on a PDF printed from a web page, it read “yes” from the start.

What OCR changed in a 20-page scan

On October 23, 2026, we built a test scan from a 20-page text document of 5,548 words. Each page became a gray JPEG of 1700 × 2200 pixels at 200 ppi, with no text in the file. The scan weighed 8,302,321 bytes, which the tool shows as 7.9 MB.

We ran OCR PDF in Chrome on a Mac with its defaults: English, Skip existing text, Straighten skewed pages and Auto-rotate pages. The result was ready 50.7 seconds after we pressed Recognize text, upload included. The page said: “Server report: text found on 20 of 20 pages.”

Then we pulled the text out of the result and compared it with the original, word by word. Of the 5,548 words, 5,536 matched, or 99.78 percent. The misses were numbers, such as a 4, a 7 and 636, plus one word. A search for “warranty” found 243 hits, as many as in the original document.

The pages stayed pictures, all 20 in the same order. The text layer did not make the file heavier: it came back at 7.6 MB, 4 percent lighter than the scan. It carried no tags for headings or reading order, which the W3C technique adds as a separate step. Our scan was clean printed text; we did not test phone photos or handwriting.

A tilted page, and what happens to the pictures

Scans often come out crooked, so we also turned two of the pages by 3 degrees. OCR PDF set them level: measured with ImageMagick, the tilt went from 2.96 degrees to zero. All 549 words on the two pages matched, and the job took 15.8 seconds.

Straightening has a price. With Straighten skewed pages on, every page picture was saved again as JPEG, straight pages included. We then switched it and Auto-rotate pages off under More settings. The same two straight pages came back with their pictures byte for byte as scanned. So switch both off when a file must keep its original images.

When you do not need OCR

Most PDFs made on a computer already have text: a file exported from Word, or a page printed to PDF from a browser. Our 20-page document was one, printed from Chrome. OCR PDF returned it in 2.1 seconds: “This PDF already has a text layer, so nothing was added.” The note pointed to Redo OCR instead.

Redo OCR is for a text layer that is wrong, not for one that works. If search and copy already work, leave the layer alone.

To get the words out of a PDF that already has text, PDF to Text reads it in your browser, without uploading it. And a photo or drawing with no words in it gives OCR nothing to find.

Where our OCR PDF stops

OCR PDF runs on our server, not in your browser. The file is uploaded when you press Recognize text and deleted as soon as the job finishes. The result stays downloadable for one hour and is wiped from the server within about two hours.

One job takes one PDF of up to 300 MB and 500 pages. You can run 30 jobs an hour across the tools that use our server. A longer book goes in parts: our guide to scanning a book covers the camera, the parts and the three modes.

It reads any of the 15 languages in its menu, up to three at once. Pick the language of the scan in “Document language” before you press “Recognize text”: a Russian page read as English comes back as Latin letters. For an editable document instead, PDF to Word recognizes a scan in the language you pick in “OCR language” under “More settings”.

A picture or a screenshot does not need a PDF at all. Image to Text reads JPG, PNG and other images inside your browser and gives you a .txt file. OCR PDF is one of more than 200 free tools on PulsarKit, with no account and no subscription.