Thread by @conspirator0

Hypothesis: individuals tweeting organically about US politics will (consciously or subconsciously) favor material related to states to which they have ties. Bots and managed accounts, on the other hand, will follow news cycles and upcoming elections.

To test this, we downloaded tweets containing the names of 14 states from the week before each state's primary. We filtered this set to accounts that had tweets for at least 3 states and one or more of a set of political terms/hashtags (#MAGA, #Resist, etc) in their profiles.

This yielded a total of 790 accounts. Checking the set for probable bots (24/7 activity and/or 90% of tweets posted via automation tools) yielded 38 accounts. 3 were false positives; the rest we categorized as right-wing, left-wing, news or humor.

Most of the accounts do not look like part of a single network, but there is an exception. @trumpWG, @twg20CY, and @twgforceone were all created January 2016, have identical schedules, and tweet the same content - hashtagged images with "TRUMP WHITE GENOCIDE" emblazoned on them.

This table shows a sample of the activity of these three bots - each one tweets three times an hour, and the selection of hashtags applied to the images seems more or less randomized.

One final observation (for the moment) regarding this technique: unlike other projects that we've undertaken that have resulted in the detection of bots, this method appears to be neutral: it flagged automated accounts on both ends of the political spectrum.