Why can't I search or copy text in my scanned PDF

Because the pages are pictures. A scanner or a phone camera photographs each page and the PDF wraps those photographs in a document shell. It opens, it prints, it looks completely normal, but there are no characters inside it. So the search box has nothing to match, the cursor has nothing to select, and a PDF editor has nothing to click. The fix is OCR (optical character recognition), which reads the images and writes a text layer underneath the page picture.

Tell the difference in five seconds

Drag your cursor across a line, or double click a word.

If individual words highlight, the file has real text and search will work. If a single blue rectangle covers the whole page instead, you are selecting a picture, and no amount of retyping your search term will help.

There is one wrinkle worth knowing. A scan that has already been through OCR also selects as text, because the invisible text layer is what your cursor grabs. So the test tells you whether text exists, not whether the page started life as an image.

What OCR actually adds

Recognition does not convert the page into a word processor document. It looks at the image, works out where the letters are, and writes those letters into the file as an invisible layer positioned under the original page image.

That is why the result still looks like your scan rather than a retyped version of it. The visible page is your own image, recompressed on the way through but not redrawn, and the searchable part is sitting behind it. Our output is named searchable.pdf for that reason: your page images with a text layer underneath.

One thing about the page does change, because recognition needs level lines. A page that went in crooked comes back straightened, up to about 15 degrees, and a page scanned sideways comes back upright. How to make a scanned PDF searchable goes through the whole job step by step, including the cleanup passes that leave the page alone and the ones that do not.

Two details of how we run it are worth knowing.

The first is that text already in the file is never read a second time or overwritten. A page that is already real text comes back exactly as it went in. A page that mixes typed text with a scanned image is treated as the mixture it is: the typed words are left alone, and the scanned part gets its own text layer underneath, so you do not have to choose between keeping one and searching the other.

The second is that the scan is cleaned up before it is read. Crooked pages are straightened, by up to about 15 degrees, and a page scanned sideways is turned the right way up; both change the file you get back, so it comes out straight. Speckle filtering and shadow levelling are applied only to the copy the recognizer looks at, so they buy accuracy without altering how your page looks.

Make the scan searchable

Why the results are sometimes poor

OCR reads what the camera or scanner captured, not what you saw when you were standing over the page. Everything that makes a scan hard for you to read makes it harder for recognition too: low light, motion blur, a page photographed at an angle, shadows near the binding, a low resolution setting, or paper texture and noise speckling the letters.

The cleanup pass exists to take the edge off exactly those problems, but it is a cleanup step, not a rescue. No recognition engine recovers a photograph taken in bad light at an angle.

One deliberate exception looks like a bug when you meet it: a page speckled with noise is not straightened, even when it is visibly crooked. Rotating specks smears them into blobs no filter can clear afterwards, which costs more accuracy than the crookedness does, so a dirty page is cleaned and left at its original angle.

If the output is full of mistakes, rescanning usually beats rerunning. Lay the page flat, get even light with no shadow across the text, hold the camera square to the page rather than tilted, and scan at a higher setting. That single change fixes more errors than any option on our page.

A practical ordering note: if the file is also too large to send, run OCR first and compress afterwards, because compression softens the page image that recognition has to read. The guide to compressing a scanned PDF covers that trade-off in more detail.

When this won’t work

• The scan is hard for you to read. If you cannot make out a word yourself, recognition will not do better. Rescan the page instead of running it again.
• The document is longer than 60 pages. The job stops rather than reading part of it: a 61 page file comes back with “That is 61 pages. OCR up to 60 pages at a time.” Split it with Split PDF and run the parts separately.
• The language is not one of the 13 we support. Recognition needs a model for the script and language on the page. Anything outside that list needs other software.
• You wanted an editable document. OCR gives you a searchable PDF, not a reflowed file with editable paragraphs. Converting to Word is a separate job with its own trade-offs, and no conversion of a scan is perfectly faithful.
• The file cannot be uploaded. We delete the upload and the result automatically about an hour later, but that does not override a workplace, school, legal or healthcare rule that forbids third-party processing. The notes on what happens to a PDF you upload explain what to ask before using any server-side tool.