Make a Scanned PDF Searchable: OCR Without Uploading the Scan
Scan a page and your computer stores a photograph of it. The words look right on screen, but to software they are just pixels — Ctrl+F finds nothing, copy-paste gives you nothing, and no index anywhere knows what the document says. OCR (optical character recognition) fixes exactly that: it looks at the shapes, recognises letters and words, and writes a real text layer underneath the image.
Why this matters in practice
- Finding anything. Ctrl+F across a 200-page scanned contract starts working.
- Reusing the words. Quotes, addresses, clause numbers — selectable and copyable instead of retyped.
- Accessibility and indexing. Screen readers and search engines can suddenly “read” the document.
The catch nobody mentions
Most online OCR tools read your document by uploading it to their servers — which is a strange thing to do with a scan of your passport, a medical record or a client contract. Our OCR tool takes the other path: recognition runs inside your browser (the industry-standard Tesseract engine, compiled to WebAssembly), and the searchable PDF is assembled locally too. The scan never leaves your device.
Getting the best accuracy
- Straighten first. A page tilted by two degrees costs accuracy on every line. Run Deskew before OCR if the scan looks crooked.
- Scan at 300 DPI. Enough detail for the engine; more only slows it down.
- Match the language. Pick the right language in the OCR tool — recognition dictionaries differ, and “English” on a French page produces confident nonsense.
- Mind the fonts. Typewriter faces and handwriting are genuinely hard; expect corrections there.
After OCR
The output is a PDF that looks identical but carries invisible, selectable text. From there it's a normal document: search, copy, feed it to the PDF to Text tool for the raw words, or convert it to an editable format with PDF to Word.
An accuracy checklist
- Straighten the page before recognition; deskewed scans read better.
- Use about 300 DPI — more pixels help only up to the point where noise starts to hurt.
- Choose the language on the page, including the correct script.
- Search for three known details afterwards: a total, a date and a name.
- Check the printed output, not just the search result.
Languages and handwriting
Recognition works best on clean printed text in a language it supports. Mixed-language pages can confuse the language model; run separate passes if the sections are distinct. Handwriting is the hard case — printed labels around a handwritten note may be recognised while the note itself is missed. Do not treat an OCR result as a certified transcription of a handwritten record.
What to do with the text afterwards
Once the scan has a text layer, it can be searched, copied and indexed, and it can be converted to Word or extracted as text. Keep the original scan alongside the searchable version if the image is the legal record. Privacy is the other half of the decision: browser-based recognition keeps the scan on your device, while an upload-based service sends it to a server. Choose the model that matches the sensitivity of the document, not just the price.
Try it: make a scan searchable — locally, for free: the scan never leaves your device and the generated PDF is built the same way.