This latter supposition is supported by the fact that there are nonhuman accounts on Twitter. Some are anodyne or were created in service to legitimate purposes. Others were built to artificially boost news stories or to inflate follower numbers.
We often call these accounts “bots,” though we often use that term loosely. Sometimes we mean automated accounts built to share information en masse. Sometimes we mean it generally as an account that simply regurgitates talking points as part of a concerted effort, though the line between this and a committed partisan gets blurry quickly.
As awareness of the existence of bots has grown and as dislike of Trump has remained constant, the two have been rhetorically merged: Surely Trump’s followers are mostly bots. That idea was heightened once we learned that Russia worked to influence the 2016 election. There are scads of bots, and many of those bots must be Russian! That’s why Trump’s follower count is so high!
This doesn’t really make much sense as a strategy. Trump’s certainly not immune to concerns about vanity, but there’s no evidence that his follower count is inflated. As president of the United States, it’s natural that he would have a lot of followers. (Barack Obama, for example, has tens of millions more followers than Trump.) There’s not much use for Russia to create bots just to follow Trump; their success in influencing the 2016 campaign appears to have come, instead, from human-controlled accounts that posted inflammatory content.
On Wednesday morning, there was a new focus on Trump and bots when the president decided to retweet nearly 60 individual tweets focused on the endorsement of former vice president Joe Biden by the firefighter’s union. This sudden embrace of a bunch of random people spurred a familiar refrain: He’s retweeting bots!
Why? Well, look at the retweets.
I asked the tweeter, Amy Webb, a professor at New York University, why she suspected these accounts were bots.
“Look at language, date account was established,” she replied. “Usual markers are there.”
It’s not my intent to pick on Webb, but those markers and other commonly cited ones — numbers at the end of a username, a lack of a profile picture — are not good indicators that an account is a “bot."
In August, I showed how the process for setting up a new account means that users can often end up with accounts lacking a profile picture, lacking much personally identifying information and, thanks to Twitter’s username generation system, end up with a username that ends in a string of numbers. Twitter wants to make the process of joining and tweeting as low-effort as possible. The hallmarks of a bot, as commonly understood, are often just the hallmarks of new users.
New users are also given recommendations for whom to follow — recommendations that often include Trump’s account.
But let’s assess the tweets that Trump retweeted on Wednesday.
Many claimed to be firefighters or the family members of firefighters, claims that are hard to evaluate and that may be untrue. (That obviously doesn’t make them “bots.”) Only five of the accounts were created in 2019; combined, they have 426 followers. Thirty-eight of the accounts are older than Trump’s presidency. Another five accounts have more than 1,000 followers, four of which were created before 2014. (Bot accounts often follow other bot accounts to boost their follower counts and to therefore appear more legitimate.)
Now we need to get a little geeky for a second.
In order for a bot to publish to Twitter automatically, it must use what’s called an application-program interface, or API. Twitter’s API system allows programmers to create tools to publish automatically, with those tools identified by name in each tweet. My tweets generally post using a program I wrote and are stamped with the name I provided to the API.
There were no third-party tools used to tweet the messages that Trump retweeted on Wednesday morning. Instead, all of the tweets were created using five tools all created by Twitter: Twitter for iPhone, Twitter Web Client, etc. In other words, these users would have had to type the tweets into Twitter’s applications or Web page. (Only three of the tweets used the Twitter.com website to post.) That’s not an efficient way for bots to operate.
A review of the recent tweets from each account shows that the users’ past tweets were hardly ever posted using questionable third-party applications. Researcher Josh Russell, for example, pointed to oddly named applications used by two of the accounts in 2009 or 2010. Those applications appear to have been legitimate but since deleted by Twitter. Our review of the recent tweets from the other accounts doesn’t identify any consistent posting using an unusual application. (One guy set up an automatic reply through IFTTT that thanked people for following him.)
Perhaps the most significant indicator that these aren’t bots, though, is that they weren’t originally trying to interact with Trump. Trump’s retweets were of people who responded to conservative commentator Dan Bongino several days ago. While Trump has 60 million followers, Bongino has a more modest 1 million. He’s got a much lower profile than Trump, and it’s hard to see why he would be a significant focus of attempted bot engagement. Trump was retweeting people who were talking to Bongino, a group necessarily less likely to have included accounts seeking to gain Trump’s attention.
There were two accounts for which we couldn’t get additional information. One had gone private since being retweeted by Trump, probably a function of suddenly being lifted to the attention of his millions of followers. Another appears to have deleted his or her account, perhaps for the same reason.
The evidence at hand, though, is pretty clear. These 60-odd random individuals appear to have been simply random individuals who a few days ago sent a reply to a conservative commentator they follow on Twitter. Suddenly, before 6:30 a.m. on Wednesday, their thoughts supporting Trump’s argument were shared with millions of people.
And just like that, people suddenly thought they were robots.