| I follow a (very) wide spectrum of accounts on Twitter and the following seemed true for the accounts I follow which scored highly likely to be bots: - it's 50/50 accurate picking actual bots I know I follow - the more simple the reading level, the higher it scored - however, extremely heady accounts also score high - the farther right/left the account goes seems to up the score a little - accounts that quote authors frequently score very high, even if they're human behind them - everything from Chinese state media scored higher than anything - everything that was frequently highly critical of China scored extremely high - most Trumpers and Antifa I follow had very low scores (but most are high-up/origin accounts) - most news organizations from both sides scored middle range, including known propagandists and "reputable" organizations - accounts from both sides of the COVID discussion were mostly low Hard to say, but my best guess is "low effort" is the defining factor. Only exception is the pro/anti-Chinese stuff, which is definitely a mix of real people and bots. I keep pretty good tabs on the accounts I follow and I think I'm pretty good at determining who is "a real person with real feelings saying real things they really think" and bots, and from this analysis, I have zero faith that this tool has any modicum of reliability in a research setting. I think it's possibly accurate enough to say "yes, this is definitely a human", but totally fails to identify bots. Thus, any research that uses this tool as a measure of bot influence is way off, because there are way too many false positives. Which I guess is a good thing for anyone who wants to say, "look at all these people being suckered by all these bots!" which is pretty unscrupulous. YMMV |