Are third-party estimates of 'fake' Twitter followers reliable?

Twitter advises users to be skeptical of claims about fake followers for well-known users.

Outside researchers in an April study determined that bot-identification software was prone to inaccurately distinguishing between humans and bots. Twitter says commercial tools used to detect bots rely on human analysis of factors such as account name, activity and location. It describes the approach as “extremely limited," since the algorithms are initially subject to human bias.

Tools like Twitter Audit and SparkToro that claim to gauge major accounts' follower-authenticity come to substantially different conclusions regarding the same accounts. Moreover, the "random samples" these tools use aren't random, but a list of an account's 75,000 most recent followers. In the context of an account with millions or tens of millions of followers, some consider the samples insufficient.