OCR (Optical Character Recognition) is the bridge between pixels and text. A scanned page is just a photograph: selectable "characters" do not exist until OCR analyzes the shapes, matches them against trained models (modern engines like Tesseract use LSTM neural networks), and writes recognized text back onto the page as an invisible, searchable layer.
OCR accuracy depends on the source: straight, clean, 300 DPI laser-printed pages convert near-perfectly; handwriting, stamps, faxes, and skewed photos degrade quickly. OCR output is therefore always a proofreading step, not a guaranteed transcription.
For PDF workflows, OCR is the necessary detour before any meaningful conversion: a "PDF to Word" conversion of a scanned file without OCR just wraps images in a .docx. Run OCR first, then convert — and check numbers character by character, because a transposed digit is more expensive than a missed letter.