Searchable PDFs and OCR: How to Turn Scanned Documents Into Text You Can Copy

If you've ever opened a scanned contract, tried to copy a clause, and gotten nothing — or searched a 40-page PDF for a name and your reader insisted there were "no results" in a document that obviously contains it — this article is for you. The problem isn't your software. It's that your PDF has no text in it at all. It only has a picture of text. OCR is how you fix that.

This is a practical guide for anyone who works with scanned paperwork: legal teams, accountants, researchers digitizing archives, or anyone who just scanned a stack of receipts and now wants them to be useful.

Two kinds of PDFs that look identical

Open two PDFs side by side — one exported from Word, one scanned on an office copier. On screen they can look the same. Under the hood they are completely different.

• A text-layer PDF stores the actual characters. The letters are encoded as text plus font information. You can select a sentence, copy it, search it, and a screen reader can read it aloud. PDFs created by "Print to PDF," Word, or most software fall into this category.
• An image-only (scanned) PDF is a photograph of a page wrapped in a PDF container. Every page is essentially one big picture. There are no characters — just pixels that happen to be shaped like words. Select tools grab nothing, search finds nothing, and assistive technology sees a blank.

A quick test: open the file and try to drag-select a single word. If a tidy highlight snaps to the word, you have text. If you can only draw a rectangle over the area (or nothing highlights at all), it's an image.

What OCR actually does

OCR — Optical Character Recognition — looks at the picture, recognizes the shapes as letters and words, and produces machine-readable text. The pdf-edit.tech OCR tool is built on Tesseract, the long-standing open-source OCR engine, run through ocrmypdf (which coordinates Tesseract with Ghostscript).

The important and often-misunderstood part: OCR here does not redraw or replace your page. It keeps your original scan exactly as it looks and adds an invisible text layer behind the image. The characters are positioned right over the words you can see, but rendered transparently, so the page looks identical while becoming fully searchable and selectable.

That design has a real benefit. A scanned signature, a stamp, a handwritten note in the margin, the exact layout of a form — all of it is preserved pixel-for-pixel. You're not trading away fidelity for searchability. You get both.

Why the invisible-layer approach matters

Because the visual page is untouched, OCR is non-destructive. If recognition gets a word wrong, the worst case is that searching for that word fails — the document still displays correctly. A process that retypes the page is riskier: one OCR mistake there changes what the document literally says. The transparent-overlay method is the safe default, and it's what makes a "searchable PDF" trustworthy for archives and legal records.

Why this is worth doing

Search. A searchable 200-page deposition turns Ctrl+F into a research tool. Finding every mention of a date or party name goes from a 20-minute page-flip to two seconds.

Copy and paste. Quote a paragraph into an email, pull a figure into a spreadsheet, lift an address into a form. None of that is possible on an image-only file.

Accessibility. Screen readers need real text. An image-only PDF is, to a blind or low-vision user, a blank page. Adding a text layer is often the single biggest accessibility improvement you can make to a scanned document — and in many organizations it's a compliance requirement.

Long-term archiving. Searchable PDFs are the basis of usable digital archives. They also feed nicely into PDF/A, the archival standard; the same ocrmypdf-based pipeline powers conversion to PDF/A for documents you need to keep for years.

Better conversions downstream. Want to get a scan into Word? The PDF to Word converter does its best work on text-based PDFs. A scanned, image-only PDF has no text for it to extract, so the result is poor or empty. Run OCR first, then convert — that single ordering fixes most "the Word doc came out blank" complaints.

Language support

Tesseract is multilingual, and the OCR tool defaults to English (eng). If your documents are in another language — or mix languages, like an English contract with French clauses — you'll get far better accuracy by telling the engine which language to expect. Accented characters (é, ñ, ü), non-Latin scripts, and language-specific letter pairings are all recognized more reliably when the right language model is applied. A French document run as English will mangle every accented word; run as French it reads cleanly.

If you're not sure, start with the dominant language of the document. OCR quality on the wrong language is noticeably worse than on the right one.

Quality tips: garbage in, garbage out

OCR accuracy is decided mostly before you ever run it. The engine can only recognize what the scan makes legible.

Scan at the right resolution

300 DPI is the sweet spot for text. Below about 200 DPI, small type and footnotes start to break down and error rates climb. Going much above 300 DPI rarely improves accuracy for normal documents — it mostly inflates file size. If you control the scanner, set it to 300 DPI black-and-white or grayscale for text pages.

Keep scans clean and straight

• Straighten skewed pages. A page scanned at a 5-degree tilt is harder to read; rows of text drift across the recognition grid.
• Avoid shadows and glare, especially with phone-camera "scans." Even lighting beats a bright flash every time.
• Use good contrast. Crisp black text on a clean white background is ideal. Faint pencil, faded thermal-paper receipts, and low-toner printouts are the usual culprits behind bad results.
• Don't over-compress before OCR. If you've already squeezed a scan down hard, the JPEG artifacts smear letter edges and confuse the engine.

Mind the order of operations with compression

This trips people up. If you want a smaller searchable file, run OCR first, then compress. The compress tool has two modes, and the distinction matters here:

• Smart mode recompresses and downsamples the embedded images while keeping any text layer selectable — so it's safe to use after OCR.
• Rasterize mode flattens every page back into a flat JPEG for maximum size reduction, which removes the selectable text. If you compress with rasterize after OCR, you've thrown away the layer you just paid for. (And if you rasterize before OCR, you've made the image harder to read.)

So: scan clean at 300 DPI → OCR → smart compress. That sequence keeps both small size and searchable text.

Realistic expectations

OCR is excellent but not perfect. On a clean 300 DPI printed page in a supported language, accuracy is very high — typically well into the high-90s percent. On a creased, faded, or handwritten page, expect more errors. Handwriting in particular is not something to rely on; Tesseract targets printed text. For anything mission-critical, spot-check the searchable result against the original.

The takeaway

A scanned PDF that won't let you select text isn't broken — it's just a picture. OCR adds the missing text layer underneath, leaving the page looking exactly as it did while making it searchable, copyable, and accessible. Scan at 300 DPI, keep the page clean and straight, pick the right language, and run OCR before you convert or compress. Do that, and a drawer full of dead scans turns into a genuinely usable, searchable archive.

pdf-edit.tech is free to use and run by TakshaGroups. Files are processed over encrypted HTTPS and deleted immediately after your output is generated.

More PDF guides