Fakespot CTO here. That's a good observation. Our engines utilize NLP heavily conjoined with proprietary ML models to detect influenced reviews that basic statistical analysis will never (or with low levels of success) be able to infer from review data sets, unless there are very basic signals such as 100 5-star reviews posted in an hour.

Fake review authors and providers are increasingly becoming sophisticated to bypass various filters in place on Amazon and other websites so you need to be able to leverage SOTA analysis to detect them. We also don't reveal how we detect certain fake reviews due to the fact that they will use that information to their advantage to further exploit our system, which allows us to be a couple steps ahead.

Anyone interested to know more about this topic feel free to email me at saoud@fakespot.com.