How to scan documents to a searchable PDF

Scanning a paper document and turning it into a PDF used to require a scanner. Now your phone is the scanner — and so is your browser. This guide covers the full path: capture a clean image of the document, turn it into a searchable PDF, and walk away with a file you can email or upload anywhere.

The two stages, made explicit

Every "scan to searchable PDF" workflow is actually two operations stacked together:

1. Capture. Get a clean image of each page. Quality here determines everything that follows.
2. OCR. Read the text out of the image and embed it under the page in your PDF, so it's searchable and copy-pastable.

People conflate these because most apps do both. If you're doing it manually — phone camera plus a separate browser tool — knowing they're separate helps you fix problems at the right layer.

Capture: phone or scanner?

On a phone, the built-in camera works fine. Apps like Apple's Notes ("Scan Documents") and Google Drive's "Scan" feature also work and add automatic edge detection. They produce JPGs or PDFs.

For best results, capture each page like this:

• Flat surface, good light. Place the page on a contrasting background (the desk shouldn't blend with the paper). Even daylight is the friend of OCR; harsh shadows are the enemy.
• Camera directly overhead. Tilt distorts the rectangle and confuses OCR. Hold the phone parallel to the page.
• Fill the frame. Get as close as you can while keeping all corners visible. The more pixels per character, the better OCR works.
• One page per shot. Multi-page captures save time but multiply errors. Single pages are simpler.

If you have a real scanner, use it. Set the DPI to 300 — higher is overkill for OCR; lower starts hurting accuracy.

From captured images to a PDF

If your camera gave you JPGs:

1. Open the JPG to PDF tool, drop your captures in, set the page size to "Fit to image" so there's no extra white border, and save.

If you captured directly to a PDF (Notes app, Google Drive Scan, etc.), you already have a PDF. Skip to OCR.

OCR: making the text searchable

A "PDF" of a scan is just images. Until you OCR it, you can't select text, search for words, or paste a quote. The fix is the OCR PDF tool:

1. Drop your scan PDF onto the page.
2. Wait. The first run downloads a ~3 MB English language model; on a modern laptop, that's a few seconds. Subsequent runs reuse the cached model.
3. The tool renders each page, runs OCR locally in your browser, and bakes an invisible text layer behind the page image.
4. Download the result. It looks identical to the input. Open it in any reader; you can now select, search, and copy.

The output is what's called a "sandwich PDF" — the original page on top (visible), the recognised text underneath (invisible but selectable). It's the standard format every modern PDF reader knows how to use.

Tips for better OCR

Skew kills accuracy. If the page tilts more than ~5°, OCR errors multiply. Most phone scanning apps auto-deskew; if yours doesn't, a manual rotate fixes this.

Contrast matters more than resolution. A 150 DPI scan of a sharp black-on-white page outperforms a 300 DPI scan of a washed-out one. If the result is poor, try increasing your scanner's contrast setting and rescanning.

One language at a time. Tesseract.js can OCR in different languages, but the current tool uses English by default. For mixed-language documents, the result is best on the dominant language.

Tables don't always survive. OCR reads text top-to-bottom, left-to-right. Tables can come out as one long list rather than structured rows. For tables, consider snipping out the table region as a separate image and processing it in dedicated table OCR.

When OCR isn't enough

Some documents need more than OCR: handwriting (Tesseract.js does poorly on cursive), high-density text (a 4pt footnote isn't really readable by anyone), or text in unusual fonts.

For these, a cloud OCR service (Google Vision, AWS Textract) gives substantially better results — at the cost of uploading the file to a third party. The /api/ocr endpoint in our codebase is the place that connects to those services if you choose to enable them. The default in-browser path doesn't.

Frequently asked questions

Does the OCR really run in my browser?

Yes. Tesseract.js is a JavaScript/WebAssembly port of the Tesseract OCR engine. After the one-time model download, everything happens locally in your browser tab.

How long does OCR take?

About 5–10 seconds per page on a modern laptop. A long document takes proportionally longer. There's a progress indicator so you can see what's happening.

Are my files uploaded?

No. The image of the scan, the OCR processing, and the resulting searchable PDF all stay in your browser. Verifiable in DevTools → Network — you'll see the model download on first use, but no upload of your file.

Does it work in different languages?

The current tool uses the English model. If you need French, Spanish, German, or another language, drop us a line — we'll add multi-language support if there's demand.

Will it run on my phone?

Yes, on a modern mobile browser. Performance is slower than on a laptop but acceptable for short documents. iOS Safari is fast; Chrome on Android is roughly comparable.

What about handwriting?

Tesseract is trained on printed text. Handwriting recognition is hit-or-miss. For handwritten notes you really need searchable, a cloud service designed for handwriting (or a dedicated app like GoodNotes) will outperform browser OCR.

Can I edit the recognised text?

The PDF to Text tool extracts the recognised text as a .txt file you can edit. The searchable PDF itself is fixed — to edit the visible page, you'd need to re-typeset the document.

The shortest version

Capture → JPG to PDF (if needed) → OCR PDF → searchable PDF. All in your browser, no upload, about a minute per page.