NVDA screen reader reads links in my PDFs wrong

Nine times out of ten that's the PDF, not NVDA. If the file was made without tags, the links are just loose text with a clickable area drawn over them, so NVDA either reads the raw address character by character or skips the link role completely. Two things to try. Open the same file in a different reader (Acrobat Reader if you were in the browser, or the other way round) - they expose untagged links differently and one usually behaves. And if the PDFs are yours, go back to the Word file and export again with the accessibility / document structure tags option ticked in the save dialog. A properly tagged PDF reads links like a web page.