Twitter’s Suspension Bots Are Out of Control
An innocent account gets caught up in a bot purge, with no explanation why
WinObs was gone. Snapped away like a Thanos victim with no forwarding address. My long-time Twitter pal Rich Hay’s Twitter account was gray, though not quite dust.
Hay and I met nearly a decade ago at an early NASA Tweetup to celebrate and witness one of the last Space Shuttle launches. We bonded over our love of space and, later, a somewhat shared Windows expertise. Hay’s operating system insights are levels above mine—he’s a member of the Windows Insider program, Microsoft’s open-software testing program—but as someone who covers the platform, I appreciated the knowledge he brought to the topic on his website Windows Observer and the associated Twitter account WinObs.
Hay’s name had come up in association with a project I was working on, so I decided to contact him on Twitter. And that’s when I discovered that WinObs—an 11-year-old account with 22,000 followers, 500,000+ tweets, and nothing but good, clean, techy content—had been suspended.
Twitter has spent a lot of time developing extensive platform rules, spelling them out in exhaustive detail on this page. Violating any of these rules can put you on the path to account suspension, but that is usually a last resort. Violators are first warned, mostly via email. That message can include the offending tweet or thread. Sometimes the email arrives at the time of suspension, but with the promise that if the offending Tweet is removed, they could be reinstated. From Twitter’s enforcement options page:
Unless a violation is so egregious that we must immediately suspend an account, we first try to educate people about our Rules and give them a chance to correct their behavior. We show the violator the offending Tweet(s), explain which Rule was broken, and require them to remove the content before they can Tweet again.
At least that’s how it’s supposed to work. But I knew for a fact that Hay didn’t fit any of the typical Twitter troll or bot profiles. Like its creator, the WinObs account is observational, inquisitive, wry, smart, friendly, and above all else, helpful.
Over the course of the next few days, I tried repeatedly to access the WinObs account, to no avail. I also soon realized that I had no other way to contact Hay. In all the years we’d known each other, including times we connected IRL at trade shows, we’d never traded emails or phone numbers. Our frayed Twitter connection was the account suspension’s collateral damage.
More than a week later, Hay finally popped up in my DMs. WinObs was back and he had a tale to tell.
It took eight days for Hay to get his account reactivated. He suspected he was suspended not because of anything he did, but because his account was swept up by Twitter’s anti-bot algorithms. If so, it wasn’t the first time that had happened to him. Hay told me that his account was suspended without warning last fall for the better part of a day. Using Twitter’s appeal process, he was able to get it back. This time was different.
“They’re going at it like a traffic cop with a quota.”
I looked at the last tweet Hay sent before suspension and a dozen or so before that. Nothing jumped out at me as a suspension trigger. Perhaps Hay was operating in a bot or spammy way that set off the Twitter bot-catching algorithm. WinObs’ nearly half million tweets is a lot, but Hay explained that his account is a combination of automated tweets and handmade ones. What he tweets is “never duplicates, never spam,” said Hay. “It’s genuine tech related news/tidbits from the 125 or so RSS feeds I track in Feedly [a news-feed aggregation service].”
Hay puts news items in a “save for later” bin, and then they’re picked up by an If This Then That (IFTTT) app and sent to Buffer, which schedules them out roughly four times per hour. I don’t see that as any spammier than someone tweeting 10 times an hour or using scheduled Tweets in Tweetdeck—something I’ve certainly done.
In the days following the suspension, as Hay frantically tried to get the company’s assistance—or at least its attention—he started to do a little research on Twitter suspensions. Reddit is, naturally, a popular destination for people to commiserate about everything, and Twitter is no exception. There’s a vibrant Twitter subreddit, and it is filled with people talking about the origins of their Twitter suspensions.
As I looked through the subreddit, I saw many former Twitter account holders who simply didn’t understand the rules or how to comport themselves on the platform and spread foul language and abuse. They clearly deserved suspensions. But there was also an equal number of confounded Twitter users who had no idea why they were suspended. They understood the rules and knew they hadn’t broken them, but they were suspended, often without any communication from Twitter. In trying to characterize the way Twitter suspends accounts, one Redditor put it this way: “They’re going at it like a traffic cop with a quota.”
Twitter’s ongoing efforts to scrub the system clean of millions of fake, spam, and bot accounts is not new. Six years ago, Twitter estimated that 5% of its user base was fake and the Wall Street Journal reported that 20 million fake accounts were for sale. At the time, Twitter re-engineered its system to prevent users from following and unfollowing users in bulk—a fairly obvious anti-bot move. More recently, Twitter adjusted down the number of accounts you can follow in a single day from 1,000 to 400.
It’s not clear those original efforts did much, and by the 2016 elections, it was obvious that a new kind of Twitter bot was on the rise. Socialbots, often backed by sophisticated A.I., were extremely busy in the waning days of the 2016 presidential election. They could steer users and sentiment in waves that helped or hurt particular candidates.
With the understanding that the platform they built to help connect the world and encourage a global conversation was being abused in geopolitically damaging ways, Twitter took far more aggressive action. It was a slow process that, initially, didn’t bear much fruit, but in late 2018, huge swaths of Twitter bots began disappearing from the services. Many users, including yours truly, began to notice a precipitous decline in followers. It happened so fast and so effectively that even those with the largest number of followers took notice. President Trump was so upset about it that he assumed Twitter was targeting him. (There was no evidence it was.)
While there are many cases where Twitter will have humans decide, based on other users reporting bad Twitter behavior, to suspend or remove an account, most of the bot removal work is done algorithmically. Those who understood the challenge Twitter faces applauded the company’s efforts, especially as it became clear that the purge would surely damage Twitter’s quarter-to-quarter growth numbers. But this weed-whacker approach is increasingly sweeping up accounts like WinObs, which have never broken a Twitter rule, but seem to have some profile element that triggers the bot trimmer.
Sweeping up people like Hay in a bot-driven mass purge and treating them as collateral damage isn’t fair, and it isn’t smart.
Another side effect of suspending accounts at this speed and scale is that Twitter support can’t seem to keep up with notification and adjudication. Hay never received a warning notice from Twitter, and many of the Reddit posters had no idea why they were suspended. They had no idea what to do.
Hay, though, was fortunate. He plays a relatively important role on Twitter for Microsoft. He’s been a member of the Windows Insider program for five years, an Insider MVP for four, and actively helps share information inside and outside the MVP network. So after eight days of zero answers and communication from Twitter, and despite filing an appeal, Hay turned to the Windows community for help.
Dona Sarkar, who leads Microsoft’s Windows Insider Program, tweeted to TwitterSupport:
It’s been over a week since you suspended of our of most important #WindowsInsiders community leader’s account @winobs. He’s done the necessary paperwork to reinstate and still nothing. We’d really appreciate it if you could investigate.
“I was so shocked when his Twitter account got suspended,” Sarkar told me later. “He’s never been the type to get even a little rude. Some get snappy or saucy on Twitter. He never does that. He’s just a good person on Twitter.”
Sarkar’s tweet quickly generated hundreds of likes and retweets and soon WinObs was back online—albeit without so much as an explanation or an apology from Twitter. But in a situation like this, noted Sarkar, “If you don’t have a friend who has a checkmark [verified] you’re kind of screwed.”
As effective as Twitter is at removing spam and bot accounts, there are many situations where accounts exhibit spam, trolling, and worse behavior without getting suspended. Often, though not always, these are verified accounts. Clearly, a verified account is usually not a bot, but it should be held to the same rules of Twitter conduct if it acts in a manner seemingly indistinguishable from a bot.
I wanted to ask Twitter, among other things, if the banning rules are different for verified accounts. (Hay’s account is not verified.) But Twitter wouldn’t speak to me on record and simply pointed me to its pages on enforcement philosophy and policy.
Twitter’s effort to rid its service of spam, bots, and trolls is admirable, and neither Hay nor I would ever recommend they stop. But Hay and others like him are the pure, beating heart of Twitter; people who joined early when the platform barely made sense and have used it in the best way possible: to connect with like-minded people and share important information. Sweeping up people like Hay in a bot-driven mass purge and treating them as collateral damage isn’t fair and it isn’t smart. Being off Twitter for eight days reminded Hay of how much he needs the social media platform, but Twitter needs him too and should learn how to treat the WinObs account and others like it with care and respect.