How to make a scanned PDF searchable: OCR explained

Why a scan has no text

When you scan or photograph a page, the result is an image — a grid of pixels that happens to look like text to your eyes. To a computer there are no letters or words in that image at all, just light and dark dots. That is why you cannot select a sentence, copy a paragraph, or use Find to jump to a word: there is nothing to select, because the text does not exist as text.

This catches people out constantly. A PDF that came from a scanner or a phone photo looks exactly like one exported from a word processor, but the two are completely different inside. One is text laid out on a page; the other is a picture of text. Everything that makes a digital document convenient — search, copy, reflow, extraction — depends on there being real text, and a scan has none.

What OCR actually does

Optical character recognition is the technology that reads the picture the way you do. It examines the shapes in the image, matches them against the patterns of letters and words it has learned, and produces actual text from them. Modern OCR uses trained models — for Pdfect, models for English and Spanish — that recognise characters with high accuracy on clean printed material.

The clever part is what happens to that recognised text. Rather than replacing your scan, a good OCR tool adds the text as an invisible layer positioned exactly over the matching words in the image. The page still looks identical — you see your original scan — but underneath, aligned precisely with what you see, is selectable, searchable text. That is what a "searchable PDF" is.

Why searchable PDFs are worth the effort

Once a scan is searchable, it stops being a dead end. You can search a 200-page scanned report for a single figure. You can copy a quote from an old letter instead of retyping it. You can extract the whole text to reuse elsewhere. For anyone with an archive of scanned documents — a business, a researcher, a household filing cabinet turned into PDFs — OCR is the difference between a pile of images and a genuinely usable library.

How to do it privately

Here is the catch with most OCR tools: to read your document, they upload it to a server. That is a real problem, because the documents people most want to make searchable are usually the sensitive ones — contracts, medical records, financial statements, legal papers. Handing those to an anonymous service to gain searchability is a poor trade.

Pdfect runs OCR entirely in your browser using a WebAssembly engine, with the language models served from the site and cached on your device. Nothing is uploaded, so the privacy of a scanned contract is preserved even as you make it searchable. After the first run the models are cached, and OCR keeps working even with no internet connection.

Getting the most accurate result

OCR accuracy depends heavily on the quality of the scan. A few habits make a large difference:

• Scan at 150 DPI or higher — very low-resolution images give the engine too little to work with.
• Keep pages straight; heavily skewed or rotated scans reduce accuracy. Rotate them first if needed.
• Choose the correct language, or both languages for a bilingual document.
• Prefer good contrast — faint or washed-out scans are harder to read. Rescanning darker can help.
• Remember that OCR reads printed text reliably, but handwriting is a different and much harder problem.