The Algorithm

youtube Sep 24, 2021

If you're an avid and frequent YouTube user like me, you’ll be familiar with the concept of The YouTube Rabbit Hole, and the mysterious YouTube Algorithm that guides you through it. And you’ll be familiar with such exclamations as “How the hell did that end up in my recommended videos feed” and “where did the last two hours go, and what am I doing with my life”. So how does this mysterious and constantly evolving algorithm work? Well, first a bit of history…

In the period between 2005 through to 2011, YouTube was optimised for clicks and views, however this optimisation had one major flaw – the click-bait video – we’ve all seen them, “the top 10 list of cats that look like Hitler” or “this man is eating a potato, you will never believe what happens next…” that kind of nonsense, annoying, frustrating, and not a great user experience.

Post 2012 this was followed up by time optimisation, basically time spent watching each video and time spent on the platform. This led to content creators either shortening their videos to make it more likely viewers would watch them until the end or make their videos longer to increase watch time overall.

2015 through 2016 YouTube began measuring viewer satisfaction directly with user surveys and response tools such as shares, likes, and not interested buttons. Basically, the algorithm had become a lot more personal. The goal was to find the video each viewer wanted to watch, and not just a video that many other people have watched in the past.

From 2016 to present day the algorithm must be able to take care of harmful content, brand safety and demonetisation. YouTube’s size and popularity has resulted in an increasing number of content moderation issues, and what the algorithm recommends has become a serious topic not just for creators and advertisers, but governments and lawmakers also. So there is some responsibility to support a diverse range of opinions, without the spread of misinformation that could be deemed harmful, Covid 19 and 5G conspiracies for example.

Anyway, enough history. The Algorithm cares about two things – finding you the right video for you to watch, and then finding you some more videos to watch.  That’s it!

When we talk about The Algorithm, we’re talking about three related but slightly different selection or discovery systems; one that selects videos for the homepage, another that ranks the results for any given search, and a third that selects suggested videos for viewers to watch next

YouTube says that in 2021, homepage and suggested videos are usually the top sources of traffic for most channels. Except for explanation or instructional type videos, for example “how to change a tyre” or “how does the stock market work”, which often see the most traffic from a search.

How YouTube determines its homepage algorithm

Every time a person opens their YouTube app or types in, the YouTube algorithm offers up a diverse array of videos that it thinks that person might like to watch. The selection is often random because the algorithm hasn’t yet figured out what the viewer wants yet

Videos get selected for the homepage based on two types of ranking

Performance: YouTube measures performance with metrics like click-through rate, average view duration, average percentage viewed, likes, dislikes, and viewer surveys. After you upload a video the algorithm shows it to a few users on the homepage, and if it appeals to those viewers (they click on it, watch it all of it, like it, share it, etc.) then it gets offered to more and more viewers on their homepages.

Personalisation: YouTube offers videos to people that it thinks are relevant to their interests based on their past watch history. If a user likes certain topics or watches a lot of a particular channel, more of the same will be offered up. This factor is also sensitive to changes in behavior over time as a person’s interests evolve.

How YouTube determines its suggested video algorithm

When suggesting videos for people to watch next, YouTube employs slightly different considerations. After a person has watched a few videos during a visit, the algorithm has more of an idea about what a person is interested in today, so it offers up some options on the right side of the screen:

Here, in addition to performance and personalization, the algorithm is most likely to recommend, videos that are often watched together, topically related videos, and videos the user has watched in the past. YouTube is as much a search engine as it is a video platform. In fact, it is considered the second largest search engine behind Google with some 3 billion searches per month.

Fun Fact Time

  • Google acquired YouTube in 2006 for $1.65 billion.
  • -YouTube is the world’s second-most visited website, its parent company Google, is number one.
  • It’s also the world’s second-most used social platform, second to Facebook.
  • People watch more than a billion hours of video on YouTube every day.
  • 40.9% of YouTube watch time happens on mobile.
  • Average Time Spent Daily on YouTube is 18 minutes.
  • The average visitor to YouTube checks out 8.89 pages per day.
  • People watched 100 billion hours of gaming on YouTube in 2020. Minecraft was the most-watched game, with 201 billion views.
  • Total YouTube 2020 revenue was $19.7 billion.
  • Apple Inc was the biggest advertiser on YouTube in 2020 having spent $237.15 million.
  • In the U.S. alone, YouTube will make 5.56 billion dollars in advertising revenue in 2021.
  • Gangnam Style was so popular that it broke YouTube’s video counter.
  • Justin Bieber’s “Baby” video managed to generate 12 million dislikes (and counting!). #deserved


Deano is part of vBridge's amazing infrastructure team, who are responsible for keeping the lights on, and making sure your IaaS experience is a happy and productive one.