Struggles of the YouTube Algorithm
If you search for “Coronavirus” or “COVID-19” on YouTube right now, you may notice something unusual about the results. You will be greeted by a banner across the top with a link to the Government’s latest information about the virus, and every video result will be from a reputable news source discussing the virus.
This oddity is because, for the duration of this pandemic, YouTube has disabled its usual algorithm for returning and displaying COVID-19 related results in favour of showing handpicked videos from trusted sources (Newman, 2020). A noble effort to stop the spread of false information for sure, but why was it necessary to suspend this infamous algorithm in the first place?
Anyone who has spent a significant amount of time on YouTube will probably have heard their preferred Youtubers talking (usually complaining) about the “YouTube Algorithm”. The purpose of this algorithm is to judge a users’ preferences and to recommend to them a list of eye-catching videos which the system feels they might enjoy, the goal to keep the user onsite for as long as possible.
Unfortunately, this algorithm is not perfect. In an ideal world, YouTube would argue that the algorithm is the audience’s preferences, that it represents these preferences in real-time and displays the videos viewers want to see. In reality, the algorithm chases the audience’s preferences. Viewers change their preferences as fashions and trends change, forcing the algorithm to adapt. In turn, this means content creators have to reinvent their content to keep up with trends; otherwise, they may fall out of favour with the algorithm.
Without the power of the YouTube algorithm on their side, creators face a situation where their content may not be displayed to all of their followers. Without enough views, the algorithm won’t deem the content worthy of being shown to new viewers, compounding the problem. Content creators themselves have coined this phase a “death spiral”.
As you can see, content creators on YouTube must perform a very careful balancing act to maintain relevance on the platform in today’s world. They must balance their content, style and delivery to maintain and expand their viewer base. The alternative is facing burnout, an illness which has afflicted many creators in the past. This algorithm has also lead to a change in the way we, the viewers, interact with content creators.
In the early days of YouTube, the videos primarily watched by a user of the site were that of the user’s subscriptions. If you liked content by a creator, you would subscribe to their channel. YouTube would then recommend more videos by that creator for you to watch. Unfortunately, this led to a cycle whereby the rich got richer. A creator with many subscribers would keep growing, while those with fewer, gained little.
“We needed something to break the cycle, and so YouTube started doing some experiments. They would essentially change what was recommended to [a creator’s] audience.” (Muller, 2019)
Derek Muller, an educational creator behind the channel Veritasium and with 10 years’ experience on the platform, analyses this shift in the algorithm in his video, “My Video Went Viral. Here’s Why”. He explains that YouTube changed the significance of subscriptions from ‘I want to see every video by this creator’, to more of a suggestion, ‘maybe I like the kind of content this creator makes’. YouTube also shifted the focus of its algorithm from finding videos with the highest number of views obtained to the amount of ‘watch-time’ that video created (Muller, 2019).
The consequences of this change were twofold. Initially, YouTube succeeded in breaking the ‘subscribers = views’ cycle by allowing the algorithm to recommend videos by similar creators with perhaps fewer subscribers, thus promoting smaller channels. This also increased watch-time on the site, as users would see new, eye-catching videos recommended to them.
Unfortunately, the content creators quickly learned that the more eye-catching and sensational the video thumbnail, the more likely the algorithm would recommend that video, regardless of the actual quality or relevance of content. Instead of watching videos from a viewer’s regular (or even at times, trusted) sources, a user would see the video whose thumbnail could shout the loudest over the rest. There is an ongoing debate that this has led to YouTube promoting more radicalised and scandalous content, as the algorithm is prone to promoting eye-catching content which agrees with a user’s current preferences and biases. An attempt by the platform to solve one problem has only created another.
YouTube is increasingly becoming a platform for young people to learn about current events, whether that is from official news sources or vloggers providing their opinions on certain celebrity figures. The YouTube algorithm does not distinguish between fact and fiction, which has facilitated a significant rise in conspiracy theories and the circulation of fake news on the platform (Roose, 2019) (Sky News, 2020).
YouTube’s handling of the COVID-19 situation by manually vetting all videos shown is a meaningful step forward in shutting down the spread of false information, but is infeasible on a large scale; there are over 720 thousand hours of content uploaded to YouTube every day (Clement, 2019). While companies such as Instagram are experimenting with algorithms to flag photos which may contain false information (a topic which will be discussed in a future post), to do so on YouTube would be a far greater undertaking.
YouTube’s algorithm for displaying content is incredibly complex. It must keep its users entertained by recommending to them tailored content which they will find interesting and enjoy, while simultaneously curtailing the spread of false information to avoid being labelled as a radicalisation tool. It must promote the growth of its creators to maintain a steady supply of content while also offering content diverse enough to allow new creators to enter the market. The algorithm is an incredibly advanced and high-speed tool which the majority of us never even stop to consider when logging in to watch videos, although it is in a perpetual state of “Work in Progress” in its unavailing attempt to appease us all.
Works Cited
Clement, J., 2019. Hours of video uploaded to YouTube every minute as of May 2019. [Online]
Available at: https://www.statista.com/statistics/259477/hours-of-video-uploaded-to-youtube-every-minute/
[Accessed 25 March 2020].
Muller, D., 2019. My Video Went Viral. Here’s Why. [Online]
Available at: https://www.youtube.com/watch?v=fHsa9DqmId8
[Accessed 2020 March 24].
Newman, J., 2020. Google and YouTube aren’t entrusting COVID-19 to an algorithm. [Online]
Available at: https://www.fastcompany.com/90480178/half-of-uruguays-covid-19-cases-can-be-traced-to-a-single-fashion-designer
[Accessed 24 March 2020].
Roose, K., 2019. YouTube Unleashed a Conspiracy Theory Boom. Can It Be Contained?. Thee New York Times, 19 February.
Sky News, 2020. Conspiracy theory videos still being promoted despite YouTube crackdown — study. Sky, 3 March.
