When It Comes To Dealing With Fake/Bot Accounts, Twitter Is (Still) Failing

Twitter’s left hand doesn’t know what the right hand is doing

Twitter began “removing tens of millions of suspicious accounts from users’ followers” earlier this month (Jul 12th, 2018, to be precise).

Here are the accounts which were followed by Twitter’s CEO, Jack Dorsey — ~3,600 accounts as of Jul 13th, 2018 — that were most impacted during what has been dubbed the Twitter Purge (sorted by number, yellow, and percentage of followers lost, green):

Here’s similar data from the (~9K) accounts which were following my account before/after the Twitter Purge:

While many are applauding Twitter’s recent efforts to curb fake/bot accounts, I, on the other hand, have become increasingly frustrated. Twitter’s approach to dealing with fake/bot accounts continues to be far too reactive for such a critically important issue.

I have spent the past 6 months researching fake/bot Twitter accounts. From bots being used to commit ad fraud (where billable events take place off Twitter’s platform) to political propaganda bots to (fake) influencers who fraudulently misrepresent their reach/influence (and much more!).. what a long strange trip it’s been!

My interest in fake/bot Twitter accounts was driven by someone buying my Twitter account 10K fake/bot followers in Jan 2016. I ended up learning a lot about Twitter’s spam detection tools as a result (TLDR: Twitter is actually pretty good at identifying spam accounts, they simply choose only to scrub a fraction of these fake/bot accounts).

North Carolina Accounts

This past March I discovered (and tweeted about) a family of Twitter accounts which were falsely representing North Carolina and various North Carolina municipalities. Similar to the @nine_oh Twitter account (more info to follow), many of these accounts would amplify Trump/right-leaning content in a coordinated fashion:

Whereas Twitter removed the @nine_oh account a day after this thread, getting Twitter to act on the North Carolina family of accounts required tremendous effort (and time):

Between March and May, I was in touch with several Twitter execs/senior employees via Twitter DM and email; additionally, I would frequently tag them in tweets highlighting my concern around the North Carolina accounts. I know my message was heard (loud and clear!), as most DMs/emails were actually answered. Still, I wasn’t taken seriously.

Mike Farb similarly felt this was an important issue for Twitter to address and followed up with this fantastic thread in early June:

Mike’s thread, fortunately, resulted in coverage by The News & Observer (based in Raleigh; serves the greater NC Triangle area). A few days later, Twitter (finally) removed the main accounts. Worth noting, however, that Twitter did not remove hundreds of thousands of fake/bot accounts which were following said main accounts.

The accounts removed by Twitter in early June were the same accounts which were literally mapped out for them ~6 weeks prior (again, I tweeted about these accounts initially in March, and, at that time, linked several Twitter employees to the thread):

To be clear, the issue with the family of North Carolina accounts has nothing to do with them amplifying Trump/right-leaning content. In fact, empowering users/accounts to amplify content is, in part, why Twitter’s platform is so powerful/special.

The problem is many of those accounts appeared to be far more influential than they actually were, as many of their followers were comprised of fake/bot/inactive accounts. To illustrate that point, here’s a tweet poll from the @greensboro_nc account, which drove only 19 votes in 18 hours (alarmingly low considering the account reflected having more than 250K followers at that time):

The person/people operating the North Carolina accounts is/are quite savvy re: social media. For example, they would frequently change @greensboro_nc’s display name to “#SocialMedia NC NC” (from “Greensboro, NC”) and were able to penetrate various groups of social media influencers (and even landed atop rise.global’s Online Influencers leaderboard):

In other words, despite the vast majority of @greensboro_nc’s followers being comprised of fake/bot/inactive accounts, they were also quite successful in driving real accounts to follow. As you can see below, 275 of my Twitter followers were following @greensboro_nc, an account that was falsely representing Greensboro, North Carolina:

India: Where Manipulated Twitter Polls Run Rampant

Two months ago I discovered a family of Indian polling accounts. The group even had several verified accounts as part of its network. Effectively fake/bot accounts were being used to manipulate tweet polls (pushing a pro-Narendra Modi/BJP agenda).

I tweeted about the operation on Jun 30th (after observing it for a month):

A few days later some of the main accounts were suspended by Twitter (coverage from The Indian Express).

Similar to the North Carolina family of accounts, I immediately notified several Twitter execs/senior employees when I initially discovered these shenanigans (and prior to the above tweet from Jun 30th). As was the case previously, Twitter simply didn’t seem to have any sense of urgency.

Turns out several Twitter India employees (including @Raheelk) had been warned about @MyVoteToday over the years; additionally, Alt News did a nice job exposing MyVoteToday back in May 2017:

For those that didn’t closely review the tweets, it’s important to highlight that MyVoteToday would frequently promote their (vile) poll tweets.

Was Twitter India knowingly allowing them to promote manipulated polls simply to collect advertising revenue? Jack Dorsey talks a lot about transparency. Twitter should have to disclose how much advertising revenue was earned via MyVoteToday-affiliated accounts to-date.

Moreover, in deleting the accounts, Twitter is deleting evidence (as is also the case with the North Carolina accounts).

There’s much more to this India/MyVoteToday story, but I’ll save that for a separate post!

Elections Integrity

Last week it was announced that Twitter will be (further) delaying revising their account verification process. According to Twitter’s product lead, Kayvon Beykpour, his team will be pausing their work on retooling verification as the task “isn’t a top priority” for them currently.

Here’s an excerpt from an email that Beykpour sent to Twitter’s health leadership team: “ I don’t believe we have the bandwidth to address [account verification] holistically without coming at the cost of other priorities and distracting the team.. We’re already doing way too many things, and focusing on this now will slow us down and lessen the quality of more important areas like elections integrity.”

If elections integrity is such an important area, why does Twitter continually ignore the tips/direction provided by researchers on this front?

There is no excuse for Twitter taking so long to act re: the North Carolina and MyVoteToday family of accounts. Both instances involved accounts which, over the years, had been reported to Twitter on multiple occasions. Both instances involved the use of fake/bot accounts (both for social proof and amplification purposes) to spread political propaganda.

While it’s great to see Twitter say they’ll be taking elections integrity more seriously, ultimately those are just words; moreover, they’re words which don’t align with Twitter’s previous actions.

Point Of Clarification

The recent Twitter Purge is most certainly a step in the right direction, but it’s the tip of the iceberg. It’s a shame that it took Twitter so long to even take this step, and I suspect it had lots to due with being pushed by Unilever’s Keith Weed.

Many people, it seems, have misinterpreted the Twitter Purge to mean that Twitter has removed all fake/bot/inactive accounts. That is not what took place. Twitter simply removed suspicious accounts (which were locked) from Followers count. Some of those, no doubt, are fake/bot accounts. Those accounts still appear in Following count, and have not been suspended/removed.

Geoff De Weaver

I began this post by highlighting a number of accounts which were materially impacted by the Twitter Purge. One of those accounts was @geoff_deweaver, having lost ~430K followers (54% of the account’s followers):

Out of the account’s last 3.2K tweets, 2,079 (65%) were retweets. Here are some of @geoff_deweaver’s retweets — descending from most retweeted — and the accounts which were most frequently retweeted:

Here’s a sampling of recent tweets (excluding retweets) from @geoff_deweaver where hashtags were prevalent and a second set of tweets where “Trump” was the first word:

In the North Carolina family of accounts section, I mentioned the @nine_oh account (as a reminder: it was amplifying Trump/right-leaning content and was removed by Twitter a day after this thread).

The @nine_oh account, the family of North Carolina accounts, and the @geoff_deweaver account are all connected (going as far back as Dec 2013):

Twitter: please start doing your job more effectively. Given your track record, I have zero confidence in your statement that elections integrity is a top priority

Follow me on Twitter (@geoffgolberg) and please share some claps (you can share up to 50!) if you enjoyed this post!

CEO & Founder, Social Forensics | Previously: Co-Founder, Elementus | Featured in BBC, CNN, BuzzFeed, and Quartz, among others | SocialForensics.com