If you cannot select text in the PDF, no conversion will give you editable text.
Checking only page one hides font problems that affect some pages.
Treat whatever comes out as a draft and plan to correct characters.
A PDF does not have to store which character it is. It can store the shape: a subset font where the glyph drawn for one code happens to look like a particular character, with no record anywhere of what that character was. Extraction then returns whatever the code maps to, which is why the output is nonsense rather than slightly wrong characters.
Text PDFs with properly embedded fonts convert cleanly. Subset fonts with a broken or missing ToUnicode table produce garbled output, sometimes recoverable by a converter that guesses the mapping. Scans have no text layer at all, so nothing can be extracted and the result is images inside a Word file, which is not a bug.
Open the PDF and try to select a sentence. If selection highlights characters and the copy pastes into a text editor, there is a text layer and conversion should mostly work. If dragging selects a whole page as one block, or selects nothing, the page is an image and no converter will produce editable text.
For a scan the only route is OCR, and Chinese OCR needs a real engine plus a review pass: budget time for correcting characters, not for converting. For a text PDF with a broken font table, try one other converter; if two fail the mapping is genuinely absent, and retyping those pages is faster than fighting it.
Same reason: the file stores shapes without a character mapping.
Sometimes. Broken font tables are guessable; absent ones are not.
Roughly. Tables and multi-column pages are the usual casualties.