How Does the YouTube Algorithm Work: A Peek into YouTube’s Algorithm Changes

How the YouTube Algorithm Works

You might not think of YouTube as an example of social media, at first glance. You are far less likely to link up with friends, families, and even customers on LinkedIn, than you are on social networks, such as Facebook or Instagram. But at its heart, it’s a social network, where you share video with other YouTubers. And like all such platforms, YouTube uses a complex algorithm to decide the position of videos in its recommendations and lists.


How the YouTube Algorithm Works:


People Watch What YouTube Tells Them To

It may surprise you to know that most viewings on YouTube occur as a result of the platform’s recommendations. Like Netflix, YouTube uses AI to determine the “best” videos for viewers (or at least the person whose account is currently logged in).

YouTube Chief Product Officer Neal Mohan admitted at last year’s CES that more than 70 percent of the videos you watch on YouTube, is due to suggestions made by one of the service's AI-driven recommendations.

This is particularly relevant to mobile viewers who devote more than 60 minutes to each viewing session. With such lengthy viewing sessions, YouTube’s AI gets the opportunity to place many potential videos to watch in front of the average viewer.


YouTube’s Algorithm has Changed Over Time

Before 2012, YouTube ranked its videos by view count – the more people watched a video, the more it would be presented to other viewers. The problem was that people learned to game this algorithm easily. All they had to do was to give a video a clickbait title, encouraging people to open it and watch a small portion of it. Of course, they would quickly realize that the video didn’t do what it promised, so they immediately left it and moved onto the next video in the queue. Before long people were complaining about the numerous click-bait videos.

YouTube changed its algorithm in 2012, this time favoring duration – watch time – and session time (overall time spent on the platform). Before long “gamers” learned the way to beat the system was to create long, drawn-out videos, that met their basic premise, but in a waffling and padded way.

The problem with needing to make longer videos, while keeping up overall session time, was that YouTubers began to burn out, with demand for tremendous amounts of content. The creators still had to make as many videos as previously, but with less time and resources per video.

It was during this era that a few new genres of video became popular – they had the potential to be long videos but were easy to make. Perhaps the most noticeable of these were video gamers who began to film themselves playing. It is much easier for a gamer to shoot himself in action during a game than it is doing a stand-alone game review, requiring planning and a script, for instance.

In 2016, however, YouTube added AI and machine learning to its algorithm. This has dramatically changed the types of videos it has served up to people. It has been criticized for highlighting conspiracy theories and fake news.

The most recent (2019) changes to the algorithm are in response to some of the inappropriate material that people upload to the platform. YouTube recently announced that they have modified the algorithm to ban “borderline content.”


The Aim of YouTube’s Algorithm

YouTube’s engineers describe the algorithm as one of the “largest scale and most sophisticated industrial recommendation systems in existence.”

Not all engineers are so polite about the algorithm, however. Guillaume Chaslot, a 36-year-old French computer programmer with a Ph.D. in artificial intelligence, was one of the engineers who created the algorithm. He became disillusioned during his three years working on the project. He told the Guardian, “YouTube is something that looks like reality, but it is distorted to make you spend more time online. The recommendation algorithm is not optimizing for what is truthful, or balanced, or healthy for democracy.”

Chaslot added that the algorithm never stays the same. It’s continually changing, placing different weightings on different signals. The engineers were responsible for experimenting with changes to the algorithm that would increase adverting revenues and the time that people spend watching videos. According to Chaslot, “Watch time was the priority. Everything else was considered a distraction.”

Ultimately, though, the system has two aims:

  1. to help viewers find the videos they want to watch
  2. to maximize long-term viewer engagement and satisfaction

The algorithm affects the six places you find video recommendations on YouTube:

  1. In search results
  2. In the recommended streams
  3. On the YouTube homepage
  4. In trending streams
  5. In channel subscriptions
  6. In notifications

How the YouTube Algorithm Works

As with most AI systems, the YouTube AI is sophisticated, and YouTube has released only limited information about it. They did publish a white paper in 2016, (Deep Neural Networks for YouTube Recommendations) which clarified some of the details (and obviously people have gained some indications from ex-employees). However, most information on the algorithm is still very secret.

We know from the white paper that it uses AI to track viewers’ perceived satisfaction to create an addictive, personalized stream of recommendations, i.e., it works to determine how satisfied/happy a viewer is with each video they play, and then tailor future recommendations to try and increase this level of satisfaction.

There are effectively two neural networks in use. The first filters videos to decide what would make a good match for the viewer’s “Next Up “recommendations.  The second neural network gives each video a score based on a range of factors (not yet publicly known), but it appears to include an allowance for a video’s newness and the frequency of uploads on the channel that uploaded this video.

The algorithm isn’t some form of movie rater. It’s not designed to determine some scale of “goodness” for videos. It’s intended to suggest videos that the particular viewer would watch. So, if somebody has demonstrated a love for B-grade movies, such as the infamous Bela Lugosi’s Plan 9 From Outer Space (deemed to be so bad that you can’t stop watching it) or the anti-drug movie Reefer Madness (so over the top that you couldn’t take a thing in it seriously), then you would find that YouTube’s algorithm would continue to recommend similar B-grade material.


YouTube Search Results

Many factors affect YouTube search results. Most of these are still unknown.

However, we know that two apparent factors have an impact:

  • How close the metadata connected to a video is to a search query term. The metadata includes things like the video’s title, description, and keywords
  • How the footage has performed to date. What types of reaction have there been to the video – likes, comments, watch times, etc.

YouTube makes it clear, though, that their algorithm confers much more than this. YouTube won’t serve you up naturally with the most watched videos on a particular topic.


Other Factors Affecting the YouTube Algorithm

According to YouTube, there is a range of different factors that have an impact on the videos that the algorithm chooses for any individual. These include:

  • The type of content that a viewer regularly watches (and the types that he/she rarely watch). If somebody spends 95% of their YouTube time watching music videos, the algorithm will predominantly serve them other music videos
  • The length of time that people tend to watch a particular video; do most people view it to the end, or do they drop out after only a couple of seconds
  • The speed at which a video becomes popular (or not). There is probably some leeway when a video is first uploaded to give it time to gain a reputation
  • How often the uploading channel creates a new video
  • The session time that people spend on YouTube
  • Engagement – likes, shares, dislikes, numbers of comments
  • Any negative feedback

Ways to Improve the Odds of Your Video = Ranking Well with the YouTube Algorithm

  1. Use accurate and appropriate words in your video title, along with other metadata. But don’t just use keywords for the sake of using it. You need to make them readable by humans
  2. Create an intriguing and compelling description for your video – that ties in with your chosen keywords and metadata phrases you have used
  3. Transcribe your video – quite a few people watch YouTube in a situation where they can’t have sound. They rely on accurate subtitles to make sense out of a video
  4. Create custom thumbnails (look carefully at your video editing software, and this may include this as a feature). A surprisingly large number of viewers pick a video simply because of an interesting looking thumbnail image. Don’t rely on an auto-generated one – you are far more likely to select the best bits yourself.
  5. Try to create a video that’s interesting enough to keep viewers right through until the end. You also want to include Calls to Action near the end of your videos to help direct the viewers to another one of your videos. For example, you can use cards, watermarks, and end screens with clickable links to try and direct viewers to go the way you want them to.
  6. Include a section in each video encouraging people to subscribe to your channel. The more subscribers you have, the more organic viewers you should have for your videos. Subscribers who allow notifications will also receive an alert whenever you upload a new video.
  7. Promote, promote, promote. Promote your YouTube videos at every opportunity. Don’t forget that you can cross-promote between your social platforms. So, for example, if you have videos on YouTube, promote them during your TikTok, Twitch, or Facebook livestreams.

Subscribe to our newsletter and get social media resources send to your inbox.

Do NOT follow this link or you will be banned from the site!