Scanned PDF has no text? How to OCR it

A digital PDF often already carries a text layer. You can select words in a normal reader. A scanned PDF is different: each page is really an image of paper from a flatbed, a sheet feeder, or a phone document mode. Until OCR runs, there is nothing to highlight.

Screenshot2Text is built for the scanned case. Every page is drawn in this tab, then read as pixels with open-source OCR. That means even a digital PDF is treated as a visual pass rather than a perfect dump of an embedded text layer. If you only need the words from a scan, that is the right behavior. If you need a searchable PDF with the original layout intact, you need a different kind of tool. This site returns text in an editable box, not a new PDF with an invisible layer written back in.

Multi-page files are processed one page at a time so a long document does not freeze the browser. The status line names the current page. A page that fails is skipped while the rest continue. Stopping a run keeps every line found up to that point.

Open the PDF in your usual reader and try three checks. Drag across a line: do the words highlight one by one, or does the whole page select as a single box? Type a word you can see into the search box: does it find it? Copy a sentence and paste it into a plain note: do you get the same words, or symbols and wrong letters?

Nothing highlights and search finds nothing: the page is an image. Either it was scanned from paper, or it was flattened or printed to an image, which turns each page into a picture even if it started as normal text. There is no text layer to copy in either case. OCR is the fix. Drop the file on PDF to text and each page is read as pixels.

We checked this with a test file: a one-page PDF holding only a picture of a three-line invoice note. A text extraction tool found zero words in it. Screenshot2Text returned all 20 words, and every line matched the original.

Words highlight, but search misses them or the paste comes out as symbols: the PDF has a text layer, but its fonts do not map the shapes back to real letters. The reader can draw the page, yet copying hands over the wrong characters. OCR is a good fallback here. Screenshot2Text ignores the embedded text layer and reads the rendered page instead, so the broken font mapping does not affect the result.

Words highlight, but copy does nothing or the copy option is grayed out: the owner may have set the file to block copying. Many readers show this in the document properties or security details. If so, respect that choice, or ask the owner for a copy you are allowed to use.

Words highlight, search works, and the paste is clean: the text layer is fine. A normal copy is the fastest route and you do not need OCR.

If the copy works but breaks at the end of every line, paste it into the result box on the PDF to text page and press Join lines. Removing line breaks from PDF or OCR text shows a real before and after.

Feeder skew, a shadow along the binding, a fold through a paragraph, and a low-contrast photocopy all hide letters. Lay the sheet flat when you can. Fill the frame with the type. Crop desk clutter and scanner lid chrome out of a document photo.

Resolution matters. A tiny preview export will skip punctuation. Use the original scan file rather than a chat-compressed copy. Straighten pages so lines of type are roughly horizontal before recognition starts.

Printed type on a straight-on scan converts most cleanly. Faded carbon copies, stamps over the words, tiny footnotes, and messy handwriting often come back incomplete. Expect to proofread anything legal or consequential.

When more than one page returns text, the output box prefixes each block with a header so you can see which page produced which paragraph. You can edit the box, copy everything, or download a .txt file for long runs.

Recognition stays in the browser. We do not upload the PDF for OCR and we do not store the document or the extracted text on a server. The first PDF run may fetch a small page-rendering script to your device; that request does not include your document.

If you have a single scanned page as a PNG or JPG instead of a PDF, the scan to text page is a closer entrance. Casual whiteboard photos belong on photo to text. The underlying reader accepts all of those formats.

Very large PDFs are limited by your computer's memory, not by a site quota. There is no account and no daily page cap, because the work runs on your machine.

Languages other than English need the matching language selected before you start. Mixed-language pages are harder; pick the dominant language or split the document.

OCR is a best-effort read. Check the output before you rely on it for anything that matters. The terms page states the as-is limit in plain language.