In the Twitterverse, a Twitter user’s follower count is cachet.
From employers deciding who to hire to customers choosing businesses to news consumers searching for authoritative journalists to celebrities negotiating endorsement fees, one’s Twitter follower count, sadly, is looked at as a status marker.
With many perceiving prestige by how many followers someone has, people attempting to game the system was an inevitable consequence. A New York Times expose from late January — The Follower Factory — pulled back the curtain more than any previous report on how lucrative shadowy businesses have sprung up in the past decade to give artificial boosts to social media accounts in order to inflate their brand and influence. The NYT article also revealed the identity of some of these customers buying phony followers, likes, views, retweets, etc. from these companies, making the companies millions of dollars. The journalists also uncovered “large-scale social identity theft” where at least 55,000 real Twitters users’ “names, profile pictures, hometowns and other personal details” were used to create look-alike bot accounts. Customers of the shadowy companies range from individuals purchasing the fake followers themselves or family members, friends, assistants, associates and agents buying them on the owner of the account’s behalf. (Separate of the NYT report, I’ve even heard stories of political operatives buying fake followers for their enemies, as anyone can buy fake followers for another’s account, in a game of subterfuge to smear their opponent.)
The NYT journalists focused particularly on one such company, Devumi, which sold over 200,000 clients — e.g. sports stars, musicians, politicians, news outlets, journalists, comedians, DJs — phony likes, retweets, and followers to artificially inflate the brand of individuals’ social media accounts. The NYT story exposed about 100 Devumi clients, mostly Americans, but also a few British. In all probability there were Canadian clients in that large trove of business records. Yet, when I reached out to two of the four authors of the NYT expose, explaining I wanted to shine some light on the Canadian angle of their story, they declined to share any information about the 200,000 clients and how many are Canadian.
Without smoking gun evidence like business records, I decided to analyze popular Canadian Twitter accounts using Twitter Audit, software that uses a complex algorithm that relies on probability to determine real-to-fake follower ratios of Twitter accounts. I obtained professional audit results for Twitter accounts of prominent Canadian journalists, comedians, politicians, personalities and media outlets. The audits were performed with a gold account, a top-tier paid account that scans all of an audited account’s followers instead of the free Twitter Audit account that only takes a sample size of up to 5,000 followers, which means it would be less accurate when looking at bigger accounts.
Twitter Audit looks at a range of factors to determine if an account is low quality, which determines whether or not the the account is classified as a bot (an automated account programmed to do tasks like tweet, retweet, like and follow).
“This score is based on number of tweets, date of the last tweet, and ratio of followers to friends,” reads an info box at the bottom of Twitter Audit’s website. “We use these scores to determine whether any given user is real or fake. Of course, this scoring method is not perfect but it is a good way to tell if someone with lots of followers is likely to have increased their follower count by inorganic, fraudulent, or dishonest means.”
The software is far from foolproof, but it does give an idea of just how real a following a Twitter user has and is a good starting point for finding out which accounts to further analyze.
Before presenting the data, here’s a brief interview via email with one of Twitter Audits founders, David Caplan, to get further explanation on the accuracy of Twitter Audit.
1) Could you further explain what the categories of 0-5 in the “real points” chart mean?
The real points are part of our algorithm which gives each account a number of points that add up to a score. The score basically indicates whether an account is identified as real or fake. More points means that the account is “more real”.
2) Is it possible that someone could have an audit score showing the majority of their followers are fake but it’s actually just that the account was promoted by Twitter (perhaps paid for by their employer) to new Canadian users, many of whom have now left the platform and have dormant accounts?
Yes, this is definitely possible and has been a theory of mine for a while. TwitterAudit can sometimes incorrectly identify inactive people as fake, but we do everything we can to prevent that from happening by tuning the algorithm and by learning from our false positives.
3) I noticed one top account in a previous audit came up with only 35 per cent real followers a few years ago, but he now has about 80 per cent real followers despite his total number of followers staying at about the same number. Is it a common practice for some users to buy a lot of fake followers and as this show of ostensible popularity attracts more real followers the fake followers are slowly shed?
Yes, many people purchase fake followers to kickstart their own following. People tend to follow people that have many followers, so you get the idea. The fake followers are usually never shed, unless Twitter removes them (Twitter will remove spam accounts and other accounts with suspicious activity). We have also updated our algorithms several times over the past few years, which can also account for changes like this.
4) How do popular accounts attract a lot of fake followers unintentionally? Can this explain some accounts with 30, 40, 50 per cent fake follower ratios?
My theory about this is that the ratio of real/fake followers on any large account becomes proportional to the overall ratio on Twitter itself. Based on the “law of large numbers”, in a way. I think that generally these kinds of accounts are promoted to new users, many of which become inactive. Generally I would say that for a very large account (1M+ followers), any score above 45% real is fine.
5) At the bottom of your front page you have a section explaining how the audit works. It says that the scoring method isn’t perfect but a good way to gauge if a user’s follower count was obtained inorganically and/or fraudulently. I was wondering if Twitter Audit has suggestions for digging deeper into some of these accounts with high fake account numbers? Is there a good way to prove that these accounts bought fake accounts in bulk?
I would wager that most fake followers are purchased by people that don’t own the targeted account. You can go buy fake followers for any account that exists. Often times we hear that people hire “social media experts” who promise to boost their follower numbers, who end up just buying fake followers. Other times, people are targeted with fake or spam followers. It is really hard to prove that someone bought their own fake followers. This article by a friend of mine (Geoff Golberg) highlights this point: https://medium.com/@geoffgolberg/when-bots-attack-af7f9f87b612
The recent NYTimes article also goes into some detail. There are ways to find patterns in followers for specific accounts if you dive in deep enough.
I intend to dive deep enough and am currently corresponding with some Twitter experts that are taking a closer look at some of the Canadian accounts audited below. This is the first installment in a series of stories about Twitter and Facebook in a Canadian context. Below are the Twitter Audit results, followed by responses from some users’ representatives, and finally some analysis of the results.
Responses to the Results
Premier Kathleen Wynne’s Office did not respond to a request for comment on Wynne’s Twitter Audit score of 49 per cent real followers.
“The NDP has never bought followers for our social media accounts. What we love most about [Twitter] is that it gives us the opportunity to hear directly from Ontarians,” said Ontario NDP media relations officer Rebecca Elming in regards to leader Andrea Horwath’s Twitter Audit score of 40 per cent fake followers.
“Mr. Singh’s twitter account is public and available to all and that is how it will remain. We do not filter or moderate who is or isn’t allowed to follow his social media accounts,” wrote federal NDP leader Jagmeet Singh’s press secretary James Smith. “In fact, such an attempt by a political leader could have inadvertent and troubling ramifications. Canadians have good reasons to be concerned about the propagation of fake news and the implications on our democracy. That is why the NDP has advocated for greater scrutiny of social media platforms especially where Canadians’ private data is concerned.”
Last month, the NDP successfully passed motion to launch an immediate inquiry into the scandal surrounding Facebook and Cambridge Analytica. In the coming days the NDP will table a Digital Bill of Rights in the House of Commons which includes universal and affordable access to the internet, enhanced privacy rights, ownership of personal data, and other key aspects of digital citizenship.”
“It’s important to distinguish between accounts managed by CBC (CBC Montréal, CBC Vancouver, CBC Calgary, CBC Ottawa, CBC News, CBC Alerts, CBC, CBC Politics), accounts managed by external production companies and the accounts of individuals who work or have worked at CBC in the past,” said CBC/Radio-Canada’s Emma Bédard.
“While CBC journalists/personalities may use their accounts for professional reasons, they are not managed by CBC so we can’t offer much insight into their individual practices. The same goes for accounts managed by external production companies.
“That said, I can confirm that that CBC has never purchased followers for any of the accounts we manage. However CBC has leveraged social media platforms (including Twitter) for advertising and promotion of our content and products, and so by extension that could be considered as promoting our social media accounts but our focus in these spaces is always on building meaningful engagement with our audiences.”
“I’ve often been tempted to change my twitter bio to ‘more fake followers than Jesus.’ I started realizing this issue more than a year ago,” said journalist and minority rights activist Desmond Cole in a Twitter direct message. “My number of followers didn’t seem realistic, it was escalating too quickly. I noticed that a lot of the bots that follow me are also following prominent Canadian journalists and media types. When I first realized this, I started to delete some of them. But the bots actually follow me faster than I can delete them. I estimate that I get several hundred fake follows a week. I’ve often wondered if they’re all coming from the same place since their profiles are so similar: few or no followers, accounts created within the month they follow me, haven’t tweeted or have only tweeted a couple of times (often retweets or gibberish), follow other Canadian media people. I also suspect the bots have a way of following you that doesn’t register on your account. You don’t get the usual ‘so-and-so followed you,’ but when you look in your follower list they’re just kinda there. Ultimately it sucks to know I’m 45% less popular than it appears. My larger worry is that the bots become self-aware and realize they’re all being used. We all know how that ends.”
“The Globe does not purchase twitter followers for its accounts. Twitter accounts for our staff belong to those individuals. The numbers you’re seeing, if they are accurate, may be because fake bots tend to link to accounts with higher numbers of legitimate followers,” wrote Globe and Mail communications manager Lanna Crucefix.
“I’ll be honest, I over did it. I’ll be the first one to admit it. A lot of people add five thousand so they can look a little bigger, a bit more attractive,” says Toronto political activist Bobby Armstrong. “It’s like putting yourself on steroids. If you look big, people will follow you, and if you’re smart and someone follows you with a reasonably good account you follow them back. Then you build yourself up, that’s the whole purpose of it.”
Armstrong credits his buying of followers to helping him build up his real following to tens of thousands and says that from time to time Twitter has rolled back his follower count by purging fake accounts.
BuzzFeed Canada did not respond to a request for comment on the 49 per cent fake follower Twitter Audit score its account received.
Twitter Canada’s head of communications Cameron Gordon (no relation to the author) provided some links to Twitter’s policies and blog posts when asked about Twitter’s bot and fake follower issues. Below is his response, which I will go through in an upcoming installment.
3-This is from March 1, 2018 and worth having a quick peek at: Twitter Health Metrics Proposal Submission https://blog.twitter.com/official/en_us/topics/company/2018/twitter-health-metrics-proposal-submission.html
Analysis of Results
When looking at the high fake follower scores of high-profile Canadians like Peter Mansbridge, Margaret Atwood, George Stroumboulopoulos, Rosemary Barton, Chantal Hébert, the band Hedley, Rick Mercer, Jagmeet Singh, Andrea Horwath, Justin Trudeau, Chris Hadfield, Tabatha Southey, Terry Milewski, Kady O’Malley, Nick Kypreos, Rob Carrick and Susan Delacourt, as well as news outlet Twitter accounts like BuzzFeed Canada, Globe and Mail, CP24, and various CBC accounts, it’s important to note that there are several reasons why they might have accumulated high numbers of supposedly fake followers.
First off, many of these audited Twitter accounts are suggested to new Canadian Twitter users as recommended accounts to follow. Many new Twitter accounts are created and shortly thereafter forgotten and abandoned by their creators. Some former Twitter users have left the social media platform but haven’t deleted their dormant accounts. Both of these types of real accounts could register as fake accounts under Twitter Audit’s metrics. So many of the fake accounts from these audits could actually be real people’s dormant accounts.
Bots tend to glom onto popular accounts. Many of the audited accounts could inadvertently magnetize thousands of bots because of their prominence on the platform. It would be difficult for these account holders to weed out these bots one by one if they’re attracting high levels like Cole. As Twitter Audit founder David Caplan pointed out, the law of large numbers means that accounts with over a million followers have a high probability of unintentionally accumulating high fake follower counts due to Twitter having a high proportion of bots/dormant accounts infesting the social media website. Some analysts believe that around 15 per cent of all active Twitter accounts (48 out of the 319 million active monthly users from this time last year) are bots, although Twitter disputes that estimation.
Anyone can buy thousands of fake followers for any Twitter account they want. For example, in a tweet included earlier in this report, Toronto Star columnist Chantal Hébert noted she had been targeted to receive a deluge of 10,000 new followers to her account overnight back in 2013.
Finally, with Devumi — just one on a list of companies — selling fraudulent social media services to over 200,000 clients, in all likelihood there are Canadians on that list. It’s unlikely the purchasing of fake accounts is a widespread practice by Canada’s high-profile Twitter users, and since the platform is rife with bot and dormant accounts, there’s no reason to necessarily doubt the responses above. That being said, there were some Canadian accounts that appeared more suspicious than others and are worth further observation.
After the NYT expose on fake followers was published, Twitter apparently purged over one million followers of popular Twitter accounts. It will be interesting to see if Socialbakers’ software reveals any high-profile Canadian Twitter accounts having a sudden dip in their following count at the same time period.
In the middle of March, CBC’s news program The Weekly spent nearly seven minutes in a segment on low-quality fake/bot accounts that were used in a weak astroturf operation in the Ontario PC leadership campaign. CBC journalist Wendy Mesley focused on how four bots promoted the hashtag #CrookedChristine in reference to candidate Christine Elliot, borrowing the pejorative adjective used by Donald Trump and his supporters to describe Hillary Clinton in the last U.S. election. In all likelihood the creator of the bots was someone trying to smear Doug Ford as Trumpesque or an ardent rogue Ford supporter trying to help his campaign.
“And now it looks as if bots have landed in Canada,” said Mesley in the opening of the segment.
But bots haven’t just suddenly landed in Canada — this has been a problem for the good part of a decade.
The sudden moral panic over Cambride Analytica data mining millions of people’s accounts on Facebook is concern over personal data being exploited that was long over due. This manipulation and studying of people’s personal data has been going on since all the way back when Barrack Obama was running for president in 2008, but at that time most of the media thought it was benign because of their approval of Obama and its newness.
So far the media in Canada have been largely in an oblivious slumber over the negative and nefarious ways social media can be exploited politically. According to eMarketer data from last year, over 50 per cent of Canadians are Facebook users and 25 per cent of Canadian internet users (7.4 million) are active Twitter users. This is a huge segment of the population of this country influenced and affected by these social media platforms.
(Last election CBC actively ran a propaganda campaign on social media to create a picture filter for people’s accounts to say that they were going to vote, basically trying to influence young people to go out and vote, despite the state broadcaster failing in its duty to help inform voters by refusing to cover one of the leadership debates. No, instead the CBC wanted to encourage people to vote for the sake of voting, hopefully voting for their preferred candidate.)
Now that the Cambridge Analytica story has awoken the collective consciousness of the Western press to the insidiousness of people’s personal data being shared and used to manipulate them on social media, you would think Canadian media would be looking for Canadian angles to this story. Instead, Canada has largely focused on how the whistle blower is Canadian and regurgitating international news outlets work on how the company affected the U.S. election and Brexit referendum. Despite there now being connections between the whistle blower and the Liberal government, articles like the one in the Liberal-affiliated Walrus magazine downplay it. In a Walrus piece — The Fake News About Cambridge Analytica — the author, with some degree of truth in it, argues the Cambridge Analytica story is overblown and is only one case of many companies that are mining social media users’ data. What’s odd, however, is the connections between the whistle blower and the Liberal government is only given passing reference in parenthesis in the article. There journalist showed no curiousity in finding out how Canada and the Liberals fit into this.
Toronto news outlet CANADALAND, which I’m a regular contributor to, has been one of the only news sources in Canada pointing out how servile the media and Liberal federal government has been to big tech companies like Google and Facebook. (The same could also likely be said of the previous Harper government, although the Liberals took over at around the time governments began to seriously look at regulating these big tech companies.)
This is the first of several installments on how Twitter and Facebook are affecting politics in Canada. My former Raving Canuck post on the Trudeau government giving Facebook an ultimatum to censor “fake news” or face regulations was the most popular post I’ve ever self-published. My two CANADALAND pieces related to social media — one on a Canadian blogger getting permanently banned from Twitter after a Conservative senator reported him for calling her obscenities, the other about Ontario Proud, documenting how the third-party advertiser in the impending Ontario election has become a wildly popular Facebook group — were both very popular. There is clearly a hunger for more of this coverage.
If you want more of this work please consider supporting Raving Canuck’s journalism below.