OCR software analyses shapes on an image and matches them to letters, digits, and punctuation. Early OCR relied on hand-crafted templates; modern engines use deep-learning models trained on huge multi-language corpora and can reach 98 %+ accuracy on clean documents.
You use OCR any time you need to search inside a scanned PDF, copy text out of a photo of a whiteboard, digitise old paperwork for accounting, or extract data from an ID or receipt.
Browser-based OCR (via WebAssembly builds of Tesseract) keeps the file on your device, which matters for anything with personal or financial data.