How Does the YouTube Algorithm Work? (1/2)

Last Updated on September 25th, 2020 at 11:24 am

the YouTube algorithm

The YouTube algorithm decides what content people watch on YouTube 70% of the time. And according to Pew Research Center, 81% of American YouTube users say they regularly watch videos that the algorithm recommends to them.

If you’re a creator trying to get more YouTube views, or a brand building out your YouTube marketing strategy, the platform’s recommendation algorithm counts for a lot. So how do you optimize your channel and videos to work with it, not against it?

YouTube usually isn’t famous for being super transparent with creators or advertisers about how the proverbial sausage is made. So in this article we’re going to take a look at the history of YouTube’s priorities when it comes to helping viewers discover new videos. We’re going to explore how the algorithm works. Then all the latest YouTube algorithm changes in this year – 2020.

A brief history of the YouTube algorithm

YouTube’s first video was uploaded in 2005. Fifteen years later, users are uploading 500 hours of video to this platform in a minute.

How do 2 billion users find exactly what they want to watch? The short answer is that it’s changed over the years. But here’s the long answer, too:

2005-2012: View count (a.k.a. clicks)

For the first seven years, YouTube rewarded videos that got many clicks, rather than the ones that kept users engaged.

Obviously, this system had a tendency to show people a lot of clickbait: misleading titles and thumbnails proliferated. Users would click, but then feel tricked, probably a little annoyed, and then abandon videos partway through. Eventually, YouTube realized that their user experience was going down the drain and changed tacks.

2012: Watch time (a.k.a. view duration)

In 2012, the platform announced an update to the discovery system designed to identify the videos people actually want to watch. By prioritizing videos that hold attention throughout (as well as increasing the amount of time a user spends on the platform overall) YouTube could assure advertisers that it was providing a valuable, high-quality experience for people.

Meanwhile, YouTube was also encouraging creators to stop fussing with algorithm optimization. For example, they make shorter videos to get a higher retention rate, or make them longer to rack up more watch time.

Instead, as it still does today, YouTube encouraged people to just “make videos people want to watch.”

2016: Machine learning (a.k.a. the algorithm)

In 2016, YouTube released a whitepaper that made some big changes. In it, product engineers described the role of deep neural networks and machine learning in the platform’s recommendation system.

YouTube neural networks and machine learning infographic

(Source: Deep Neural Networks for YouTube Recommendations, 2016)

Of course, for all the impressive jargon, this whitepaper wasn’t a tell-all. You can read it, but even if you understand it (or get your smart friend to explain it to you), it’s not the equivalent of Coca-Cola’s secret recipe. It’s more like if Coca-Cola announced that the reason their beverage is so tasty is because it undergoes a carbonation process and also there is sugar in it.

At this point, we still don’t know that many details about what’s under the YouTube algorithm’s hood. But we do know that it measures viewers’ perceived satisfaction to create an addictive, personalized stream of recommendations.

2016-2020: Borderline content, demonetization and brand safety

For the past few years, YouTube has faced plenty of questions about the kind of videos its algorithm surfaces and promotes.

According to YouTube CEO Susan Wojcicki, YouTube is taking its responsibilities seriously, and trying to balance a broad, fair range of opinions with making sure that outright dangerous information doesn’t spread. For instance, YouTube says that algorithm changes in early 2019 have led to 70% less watchtime for “borderline” content. (Borderline content is defined as content that doesn’t quite violate the platform’s community guidelines, but is harmful or misleading.)

It’s a complicated issue because it touches every issue: from white supremacy to the coronavirus. For instance, in March 2020, YouTube creators say the platform was demonetizing videos that so much as alluded to the existence of the coronavirus. YouTube’s position, meanwhile, is that it wants to support a diversity of opinions (i.e., how governments should respond to the coronavirus). But not the dangerous ones (i.e., videos saying the virus is a hoax, or that drinking hand sanitizer will cure it). Wojcicki announced that “when people come to YouTube searching for coronavirus topics, on average 94% of the videos they see in the top 10 results come from high-authority channels.”

No matter where you stand, the developments are ongoing. So, this is an important discussion for both creators and advertisers to keep informed about.

If you’re a creator, remember that just because the algorithm is rewarding the content you make with high visibility and ad revenue doesn’t mean YouTube won’t turn around and demonetize your channel or video if your content crosses the line into something advertisers find objectionable.

Meanwhile, advertisers need to know that their sneaker ads aren’t funding anti-vaxxers or conspiracy theorists. The YouTube algorithm in its current form helps to demonetize borderline content, mostly to protect brands. At the same time, YouTube says says it might never be able to guarantee 100% brand safety.

How does the YouTube algorithm work in 2020?

According to YouTube, the algorithm is simply a “real-time feedback loop that made lists of video to each viewer’s different interests”. It decides which videos that YouTube will suggest to individual users.

The algorithm’s goals are twofold: find the right video for each viewer, and keep viewers watching. Therefore, the algorithm is watching user behavior as closely as it watches video performance.

The two most important places the algorithm impacts are search results and recommendation streams.

How the YouTube algorithm influences search results

Unsurprisingly, the videos you get when you search “carnivorous house plants” will be different from the videos I get when I search “carnivorous house plants.” Search results are based on factors like:

  • Your video’s metadata (title, description, keywords) and how well those match the user’s query
  • Your video’s engagement (likes, comments, watch time)

How the YouTube algorithm influences recommended videos

The recommendation stream is a two-fold process for the algorithm.

First, it ranks videos by assigning them a score based on performance analytics data.

Second, it matches videos to people based on what they watched in the pass or their watch history. And what similar people have watched as well.

The idea is not to identify “good” videos, but to match viewers with videos that they want to watch. The end goal is that they spend as much time as possible on the platform.

For the record, there are three different places the algorithm makes a big impact:

  • Your YouTube homepage
  • Trending videos
  • Your subscriptions
  • Your notifications
Share this post