A text PDF contains a readable text layer; a scanned PDF mainly contains images. The extractor reads text layers first, uses OCR where necessary, and combines readable text.
OCR defaults to your browser or device language and supports English alongside it. Model selection is not automatic document-language detection. Other languages, handwriting, skew and low-resolution scans may be inaccurate.
Process and check
Test a few pages before processing the whole file. DOCX export reflows editable text rather than restoring PDF tables, pagination or columns. Check amounts, dates, names and identifiers carefully.