2022-01-12
Furry social media is a weird and wild place, with many, many people vying for the attention and admiration of others. Hundreds of thousands of artists, musicians, fursuit makers and other creators looking to earn a living through their work. In the digital social game, appearances and first impressions matter.
In this post, I dig into one furry's attempt to bulk up their own Twitter following through illegitimate means.
Garo Shadowscale is an anthropomorphic dragon and video game variety Twitch streamer. He also has a YouTube channel, a community Discord server and, quite importantly, a Patreon where the higest subscription tier is an eye-watering $500 USD per month. It lets you message him twice a day.
In August of 2021, a furry called Danderwoolf posted details of Garo's treatment of his fans, in an attempt to warn people of his manipulative behaviour. At the time, the post did not gain much traction.
Garoshadowscale's selfishness and narcissism is at the next level, this is 100% not ok to treat members and fans like this, they are people, not tools
— Danderwoolf (@danderwoolf) August 4, 2021
His discord server is legit like a cult as well, I made another tweet prior to this one about it
Stop #Garoshadowscale #furry pic.twitter.com/oDvN6TndqU
Then, in early November that same year, a furry by the name Leto posted screencaps from Garo's Discord server, highlighting the intensely overbearing set of rules imposed upon server members. The post went semi-viral within the furry community, and numerous other people began sharing their own, equally concerning, interactions with Garo.
this is like.. the most toxic set of rules ive ever read LOL pic.twitter.com/wWZY0or2RQ
— Leto, God-Emperor of Dune (@OmgLeto) November 3, 2021
Much of Garo's behaviour has already been covered by multiple people on both Twitter and Youtube. Danderwoolf has continued to cover the topic, as well as Lanza and Nidoqueer and Zechymoon and BetaEtaDelota, among countless others. I will not be discussing the details in this post.
Garo has indisputably acted in a horrible and harmful manner. His exploitation of others is beyond enough to justify his removal from the furry community. I don't intend for this post to be some kind of "gotcha", like this is going to finally be what kicks him off the internet. I'm not doing this to bring about change. I'm not doing this as a callout. People's opinions of him cannot sink any lower.
I'm doing this because it's funny.
Social Blade is a website which tracks metrics across multiple social websites. It's a useful tool to see other people's social media growth and behaviour over time. When the Garo drama originally started, it's one of the first things I checked. One of the things that stood out to me was the graph of the number of Twitter followers gained per month:
Most months, Garo gained between roughly 50 and 150 followers. However, in September of 2020, he gained a whopping 2000 followers. Sometimes people hit the algorithm just right and a post of theirs goes viral, but for someone's who's primarily a streamer and youtuber, you'd expect to see a correlated up-tick on other platforms. Social Blade also provides Twitch and YouTube stats, but Garo had no corresponding increase in followers or viewership for that time period.
It would be easy to simply say "oh those are fake followers", shrug and leave. But we can do better. We can do much better.
You can see the Social Blade page for Garo's Twitter here.
Side note: Some people were claiming the numbers on the Social Blade page for his YouTube were evidence of purchasing subscribers, because the daily increases were all multiples of 100. This claim is bogus. It is the result of YouTube no longer publicly publishing exact subscriber counts. Social Blade explains the situation here. Garo might be purchasing fake YouTube subscribers, but this piece of evidence does not support that claim.
Soon after Leto's tweet, Garo set his Twitter account to private, so I was not able to perform analysis in the heat of the moment. In Decmber 2021, I checked his Twitter account again, and it was unlocked. So I decided to take the opportunity to dig deeper.
Twitter provides an API - effectively a way to let computers talk to the site. It allows us to grab large amount of data - tweets, user profiles, etc - from the site and analyse them far faster than a human could manually.
First, I grabbed a list of Garo's followers. When I did this, there were 2751 of them.
A Twitter user by the name Conspirador Norteño describes Follow Order by Creation Date scatter plots. These are an extremely useful way of analysing Twitter followers, so we jump straight to producing one.
We'd like to investigate the spike in Garo's follower count we noticed on Social Blade. So we would really like to know when each of Garo's 2751 followers actually followed him. Unfortunately, Twitter doesn't give us this information directly, so we need to estimate it. Twitter tells us two things: First, the creation date of any account. For humans, it's visible on the profile page of any account, and it's also visible from the API. Secondly, Twitter gives us Garo's list of followers in the order they followed him.
This allows us to estimate when each account followed Garo, since each follower must have followed after the account creation date of all the previous followers.
We then create a scatter plot of the followers, with the x axis representing the follow order, and the y axis representing the account creation date.
There's a few observations we can make from this plot, but the most glaring feature is the horizontal line between follower number 600 and 1500. This means we have estimated that a large number of people followed Garo in a short space of time. The y-position of this line gives us the estimated follow date for these people, which lands right in the middle of September 2020. This lines up with the anomaly in the Social Blade data. We're definitely onto something.
This graph doesn't include all the accounts seen on the Social Blade data, since we can only analyse the accounts which are still following Garo. It's likely that a large number of the spurious followers have since either unfollowed Garo, or been suspended from Twitter for violation of the terms of service.
I manually looked at a handful of the accounts in the suspicious time period, and one thing stood out: They were definitely not furries. I don't have the time to personally go through all of Garo's several thousand followers, but it turns out we can get some help from elsewhere.
A furry by the name of Zenith previously developed a program to detect if a Twitter user was a Furry or not. While I believe it was originally developed mainly for entertainment purposes, we can put it to use here.
FurryDetector works by analysing the text of a user's tweets, counting the number of occurrences of certain words, then running that data through a neural net. The details of how the program works are outside the scope of this blogpost, but I want to highlight a few caveats of this technique: FurryDetector only looks at text, and doesn't analyse images at all. FurryDetector only understands English; users posting in other languages are likely to get a very low score. Finally, we cannot run analysis on locked accounts.
A later version of Zenith's FurryDetector also takes into account the "furriness" of people you follow. However, this is not suitable for our purpose, because the followership of these accounts is under question.
The Twitter API, despite allowing us to perform actions much faster than a human, still has rate limits built in, to prevent us scraping data too quickly. Scraping all the data required to give every one of Garo's followers a FurryDetector score took about an hour.
Of Garo's 2751 followers, I was unable to give 383 of them a FurryDetector score. The 2368 remaining ones we put into a histogram, and wow.
That's a lot of users with a very low score. There's definitely a cluster at the upper end, but the bulk of these users - upwards of 1000 of them - don't appear to be furries.
For comparison, I ran the analysis over my own followers (354 users with unlocked accounts), which produced some very different results
Finally, we combine the original scatter plot with the FurryDetector scores for Garo's followers. Users with a low furry score are in cyan, while those with a high score are in pink. Accounts for which I could not produce a score are denoted by black dots.
Garo's fake followers are glaringly obvious. Almost all the accounts in the suspicious time window have a critically low furry score. These are accounts that have no genuine interest in his content, purchased to bulk up his numbers and make him seem bigger than he really is. These accounts are not necessarily all "bots". Some of them appear to be real people who have had their accounts compromised in some way - potentially through password leaks or malicious services such as "Round Year Fun". But either way, these followers aren't genuine.
There is no shortage of furries on Twitter who will tell you that follower counts do not matter. They are wrong. For any kind of artist, streamer, videographer, musician or anyone else who derives their income from the internet, having a sizeable, reliable audience of (real) people is critical.
Outside of creators, having a reputation that precedes you, and access to the ears of many, can be either a blessing or a curse, depending on how you utilise it. For people who are simply on Twitter to socialise with others, rather than earn a living selling art, follower counts act as a rough proxy metric for social desirability. However, since the number is so prominently displayed on Twitter profile pages, it's also a social advertisement, telling other people how "cool and popular" you are.
Garo lives in the overlap between these two kinds of user. He exists in the toxic intersection between socialising and content production: Garo sells simulated friendship.
The illusion of being socially connected, and well liked, is part of what he uses to leverage people into his cult. While he primarily exists on other platforms, and probably doesn't recruit much through Twitter, he still maintains a presence on the site. Others have theorised about the extent of the act: That the bulk of his following is likely faked, and the rules about DMing other server members are in place to prevent people uncovering the farce. Personally, I think there are (or at least were) a decent number of real, paying subscribers caught in his web. It's possible that even Patreon subscriber numbers are bulked up by alt accounts. Given that Patreon takes a (relatively small) 10% cut, he only needs one real patron for every nine fake ones to break even, so this isn't an infeasible strategy.
There is so much data sitting out in the open, available for anyone to pick up and start analysing. There's a lot you can learn about people (or robots) if you know what to look for. Twitter is rife with botnets spreading political disinformation, but the average user has little available to them to separate real people from malicious puppets. Analysis of the kind I've done here is unfortunately a skill that cannot be picked up overnight, and Twitter themselves has done very little to combat botnets and fraudulent activity on the site. Users need better tools to protect themselves.
However, I absolutely do not want to set a precedent of adding data analysis to the furry drama cycle. It's potentially too powerful. Statistics and data analysis is a field full of pitfalls, and intentional misrepresentation of results presents an opportunity to mislead and manipulate large numbers of people. I've done my best to ensure the data presented here is accurate, but there may still be issues with it. A lot of discourse that occurs in the furry community is ultimately a social issue, and trying to solve it with a technical analysis not going to work.
Ultimately though, the purchased Twitter followers are a small footnote in Garo's biography of transgressions. He has done - and continues to do - far worse things. He doesn't deserve to wield any of form of power or influence in the furry community, and he definitely has no place extorting vulnerable fandom members for profit.