Explanation
The thing that separates short-form voiceover from every other kind is pace. Broadcast and audiobook narration is written and read at a speed that feels unbearably slow in a feed, and the single most common problem with generated voiceover is not that it sounds artificial but that it starts half a second late and never catches up.
The second difference is that the script has to be written for the ear. Subordinate clauses, parentheses and long noun phrases all survive on a page and all collapse when someone has to say them aloud at speaking pace.
The common mistake
Writing the script as text and then reading it. Anything whose meaning depends on punctuation a listener cannot see has to be restructured, because comprehension of spoken sentences falls off sharply past about fifteen words.