The PDF-to-text tool asks the local PDF library for the text already stored in the file. If a page is only a picture, the result says so. The tool does not recognize letters from a scan.
Text-to-PDF writes a new file with a built-in Latin font, including German umlauts. Characters that font cannot draw are listed and left out. The page does not claim those characters were embedded.
A DOCX import keeps the raw words, up to 5 MB, and drops layout, tables, notes and comments. None of these tools upload the file. A hash, a QR code and a subtitle conversion stay on the device as well.