Open a scanned PDF, press Ctrl+F, type a word you can plainly see on the page — and get zero results. Frustrating, but logical: a scanner does not capture text, it captures a photograph of text. Every letter is pixels, not characters, so search, selection and screen readers all fail. Optical Character Recognition (OCR) bridges that gap by reading the pixels and embedding a real text layer back into the file. Here is how it works, when it fails, and how to get results clean enough to trust.
Why scanned PDFs behave like images
A digitally created PDF (exported from Word, say) contains text objects with fonts and positions — your reader knows exactly which character sits where. A scan contains one big image per page and nothing else. That is why scanned files are larger, why you cannot select text in them, and why phone photos of documents behave identically. Any PDF that came from a scanner, photocopier or camera needs OCR before it becomes a first-class document.
How OCR works, in one minute
An OCR engine — ours uses the open-source Tesseract engine — processes each page in stages: it cleans the image (straightening skew, removing speckles), finds text regions and lines, recognizes each character, then reassembles words using a language dictionary to fix ambiguous reads (telling 'rn' from 'm', for instance). The recognized text is embedded as an invisible layer aligned over the original image, so the page looks identical but every word is now selectable and searchable. That layered format even has a name: searchable PDF, sometimes called a 'sandwich' PDF.
Getting a clean result: scan quality matters most
OCR accuracy is decided before the software ever runs. Three things dominate: resolution (300 DPI is the sweet spot — 150 is marginal, below that expect errors), skew (even a 2-degree tilt measurably hurts recognition, so use your scanner's de-skew or a flat, squared-up phone shot), and contrast (dark text on a clean background; pencil, highlighters and coffee stains are the enemy). One blank or near-blank page can even abort some OCR pipelines entirely — if a job mysteriously fails, check for empty pages first and remove them.
The fastest accuracy upgrade is free: rescan at 300 DPI, black-and-white or grayscale, with the page flat and squared. Software cannot recover what a bad scan destroyed.
Run OCR in seconds
Upload the scan to the free OCR PDF tool and run it — you get back the same pages with a searchable text layer added. Then verify: try selecting a paragraph and searching for an uncommon word. Skim the first and last pages, where scanners most often degrade. For multi-page office scans, accuracy in the high nineties is typical on clean originals; handwriting, decorative fonts and two-column layouts with tables will need manual review regardless of engine.
What to do with the searchable file
Once text is real text, the document joins your normal workflow: convert it to editable form with PDF to Word, shrink bloated scans with compress PDF (scanned pages compress very well — see how to compress without losing quality), or extract figures with the image extractor. If the scan came out sideways, fix orientation with rotation before OCR rather than after, since upright text recognizes far better. And when you need the opposite direction — a clean digital file to share as a fixed-layout document — our PDF vs Word guide explains which format to send.
Batch workflow: processing a 200-page scan
Single pages are easy; the real test is a 200-page office file. Work in this order. First, inspect the scan end to end and delete blank, upside-down and duplicate pages — one bad page can stall or corrupt a whole batch job, and removing ten junk pages up front saves more time than any setting. Second, split very large files into chapters of 30–50 pages if your tooling struggles; smaller jobs fail less often and let you verify as you go. Third, OCR each chunk, then spot-check the same three places every time: page one (scanner warm-up artifacts), any page with a table or stamp, and the final page (feeders skew the last sheet). Fix problems in the images and re-run rather than hand-correcting hundreds of words.
Searchable PDF vs plain-text export
After OCR you have a choice. A searchable PDF keeps the original page images with invisible text underneath — best for archiving, sharing and legal records, because the document looks exactly as scanned while remaining searchable. A plain-text or Word export keeps only the recognized words — best when you need to edit, quote or feed the text into other software, but formatting and signatures are lost. When in doubt, keep the searchable PDF as your master copy and export text only as a working derivative. Storage is cheap; rescanning is not.
Free vs paid OCR: when to upgrade
Free OCR handles clean printed pages brilliantly, so most people never need to pay. Consider paid software only for specific pain: large handwriting volumes, dozens of languages in one archive, or complex financial tables where cell-level accuracy pays for itself. Even then, trial the free tier first on your hardest ten pages — if it scores above 95% there, it will handle the rest. For everyone else, the money is better spent on a faster scanner than on fancier recognition.
When OCR is the wrong tool
Three honest limits. Handwriting recognition remains unreliable for anything beyond neat block capitals — retype it. Heavy mathematical notation and complex tables usually need dedicated software or manual cleanup. And OCR cannot make a document accessible on its own: a proper accessible PDF also needs tagged structure and alt text, which no one-click tool fully provides. For ordinary printed pages, though — contracts, books, receipts, government forms — modern OCR is fast, free and accurate enough to trust after a quick skim.