That's assuming it is actual people reading it, and not just click fraud. Amazon has been battling that for years, books filled with literal gibberish reach the top few hundred titles in a category and the only reasonable explanation is that some group(s) are gaming the system by buying/phishing Kindle Unlimited subscriptions then skimming through their own "books" to turn a profit.
Using LLMs to make the books vaguely coherent instead of meaningless word salad adds a layer of plausible deniability which might make Amazon slower to notice what's going on.
It seems like a kind of strange strategy for click fraud. A KU subscription is $10, Amazon takes a 30% cut, and for $700 a month you would need 100 KU bots with unique accounts and payment info.
https://www.vice.com/en/article/v7b774/ai-generated-books-of...
Using LLMs to make the books vaguely coherent instead of meaningless word salad adds a layer of plausible deniability which might make Amazon slower to notice what's going on.