I’m confused. Teachers/professors have said that using AI to detect papers written by AI is highly unreliable. How can this work effectively with a much smaller sample of text to work with (even when it looks for “similarities” between multiple reviews)? What happens in a week when Amazon starts writing fake reviews in different tones/“voices”/styles that are intentionally difficult or impossible to compare?
It's more than just bots, a lot are copy pasted 5 star reviews on shitty products. Or take for instance when sellers are allowed to completely change the listing but still have old reviews from a totally different product. Hopefully this is what they will filter out.
Fakespot used to reveal more about how they detected fakes, but as you say there are obvious issues with that, as it's a bit of an arms race. They don't just look at the text of the individual review though. Folks who buy reviews tend to get them from "review farms" that do reviews for a lot of products, and they don't have an infinite number of Amazon accounts to use for that, so there are network effects that can be powerful indicators, and that aren't easy for manipulate.