Scanned vs Digital PDFs: Why Text Extraction Sometimes Returns Nothing
Two PDFs can look pixel-for-pixel identical while being opposites inside: one contains text, the other contains a photograph of text. Every mysterious PDF failure (empty extractions, unsearchable contracts, screen readers going silent) traces back to this invisible difference.
The one-second test
Open the PDF and try to select a few words with your mouse. If a text cursor appears and words highlight individually, the file contains real text. If your selection draws a rectangle, or nothing selects at all, you are looking at an image of a page, which is what every scanner and most phone-camera apps produce.
A second tell: search the file (Ctrl+F) for a word you can see on the page. Real text is found instantly; a scan reports zero matches for a word directly in front of you. The page has no idea the word is there, in the same way a photograph of a book does not know its own story.
Why the difference exists
A born-digital PDF (exported from Word, generated by an invoicing system) carries drawing instructions that include the actual characters: the file knows a letter A sits at this position in this font. Text extraction just reads those instructions back; it is fast, exact, and preserves every character.
A scanner knows nothing about letters. It measures light and produces a grid of pixels; some pixels form letter-like shapes, but that meaning exists only in a human's eye. Extracting text from such a file honestly returns what the file contains: no text. Tools that show an empty result on a scan are not broken, they are truthful.
OCR: manufacturing text that was never there
Optical character recognition looks at the pixel shapes and GUESSES which characters they represent, then writes an invisible text layer behind the image. The page still shows the original scan; searches and selections hit the hidden layer. This is how scanned books become searchable.
Because OCR is guessing, it inherits every weakness of the input: skewed pages, coffee stains, unusual fonts, handwriting and low resolution all raise the error rate. Even at 99% accuracy, a dense page holds a couple of dozen wrong characters, which matters enormously if the page is a contract or contains IBANs. OCR output is a draft to verify, not a record to trust.
Scanning so your future self can use the file
Resolution decides everything downstream: 300 DPI is the standard that keeps OCR accurate and print crisp; 150 DPI produces smaller files that read fine on screen but push OCR toward errors; anything below that is a photograph of a memory. Scanning above 300 DPI mostly buys file size, not fidelity, for ordinary documents.
Prefer black-and-white or grayscale for text documents (color triples size and helps OCR not at all), keep pages straight since skew is OCR's worst enemy after blur, and if your scanner offers 'searchable PDF', turn it on; that is built-in OCR at scan time, when the original is still in your hand for checking. The digital original remains king: whenever a real PDF export exists, extracting from it beats scanning the printout every single time.
Try the tool:
PDF to Text