What Is OCR and How Does It Work With PDFs?
OCR (Optical Character Recognition) looks at an image of a page, recognises the letters, and adds real text to the PDF. It turns a scanned PDF into one you can search, copy and convert to Word.
The problem OCR solves
A scanned page is just a photo. To a computer, the word “Invoice” in a scan is a pattern of pixels, no different from a picture of a cat. OCR analyses those pixels and works out which characters they represent.
How OCR works (simplified)
- Clean up: straighten (deskew) the page, remove specks, increase contrast.
- Layout analysis: find columns, paragraphs, lines and tables.
- Recognition: identify each character or word, today mostly with neural networks.
- Language check: use dictionaries and language models to fix likely mistakes (“rn” vs “m”).
- Output: add an invisible text layer to the PDF, or export to Word/text.
Free ways to OCR a PDF
| Tool | Platform | Notes |
|---|---|---|
| Google Drive / Docs | Browser | Upload PDF, right-click > Open with Google Docs. Gives editable text, loses layout |
| Microsoft Word | Windows/Mac | Opening a scanned PDF in Word runs OCR automatically |
| Apple Live Text | Mac, iPhone, iPad | Select text directly in scans and photos in Preview, Photos and Files |
| Adobe Scan | iPhone/Android | Scans with the camera and OCRs automatically |
| Microsoft Lens | iPhone/Android | Scan to searchable PDF or Word |
| OCRmyPDF | Windows/Mac/Linux (free, open source) | Adds a text layer while keeping the original scan; runs locally |
Tips for better OCR results
- Scan at 300 DPI. Lower loses detail, higher rarely helps.
- Use black & white or greyscale for text documents.
- Keep pages straight and flat, with even lighting and no shadows (phone scans).
- Set the correct language in the OCR tool.
- Always proof-read numbers, names and amounts. OCR errors are most costly there.
Frequently asked questions
Is OCR 100% accurate?
No. On clean, 300 DPI printed text modern OCR is typically above 98–99% accurate. Accuracy drops with handwriting, low resolution, skewed pages, unusual fonts, stamps or poor lighting.
Can OCR read handwriting?
Modern cloud OCR (Google, Microsoft, Apple) can read neat handwriting reasonably well, but results are much less reliable than for printed text.
Does OCR change how my scan looks?
Usually not. Most tools keep the original image and place an invisible text layer behind it, so the page looks the same but becomes searchable.
Was this page helpful?
Thanks for the feedback!