What is the YouTube algorithm - What YouTube doesn`t tell
Written by: Dexxter ClarkIn short: an algorithm is a piece of computer code to perform a certain task.
In the case of YouTube: serving the viewer the most relevant videos.
Did you know that the Google Artificial Intelligence algorithms are used by YouTube?
Did you know that there are 6 different algorithms on YouTube?
In this article I help you to understand the different algorithms and what you can do to beat them.
YouTube is very secretive about the specifics of these AI (artificial intelligence) algorithms they use for YouTube.
Nobody knows exactly what is going on.
There is a common understanding on how these algorithms work because YouTube released vague descriptions called “common practices”.
There are six official algorithms that YouTube talks about in their “common practices”, let’s take a look at each of them in this article.
But first I want to dive into the Algorithms of Google, since they are a crucial part of the YouTube ecosystem.
Google’s algorithmsGoogle put its artificial intelligence algorithms, called “Google Brain”, a couple of years ago into YouTube.
The algorithms may be the same in core, but they are probably modified to fit YouTube’s needs.
They probably changed the parameters of the AI and trained it with other samples.
That does mean that there are striking similarities between the two and I wouldn’t be surprised if they were connected and update each other on a daily or hourly basis.
By looking at Google’s algorithms it sheds at least some light into the otherwise proverbial black box.
Keep in mind that AI constantly learns: what works today, won’t work tomorrow.
Cloud VisionCloud Vision is an AI algorithm that interprets images.
It detects facial expressions, text and objects.
The algorithm tries to assign labels to the image.
It also detects if the image is spoof and the likelihood of adult content.
I’ve seen a strong relationship between videos with terrible view counts and the algorithm misinterpreting the thumbnail.
Cloud Vision detects most thumbnails of my bad videos as spoof and assigned tags that had nothing to do with the video content whatsoever.
Nowadays I check my thumbnails in Cloud Vision and tweak them if it makes a mess of the tags.
When you go to https://cloud.google.com/vision/docs/drag-and-drop, you can check your own thumbnails.
So much of the thumbnail “fog” will clear up.
Cloud Video IntelligenceVideo intelligence is basically the video version of Cloud Vision.
It detects the content of every single frame in the video, probably by using Cloud Vision for every frame.
Video intelligence also does speech recognition and even can make a distinction between individuals A and B by detecting who said what.
From all data of all of the individual frames and the speech recognition combined, it determines the topic of the video.
For YouTube this data is combined with the title, thumbnail and description.
The reason why tags are pretty much obsolete nowadays is, because the “opinion” of the algorithm has much more weight than user input via tags.
You can specify a video category in the video editor in YouTube studio.
However YouTube doesn’t use this anymore, because the algorithm(s) takes care of that nowadays.
Using a lot of topic related b-roll on a “talking head” video, might help the algorithm to better determine the topic of the video. You can check out the data Cloud Video Intelligence generates with a pre defined set of videos on https://cloud.google.com/video-intelligence/ and scroll to section: “Demo van Video Intelligence API”
Natural LanguageThe natural language AI is used by Google to understand content of websites.
This is used by YouTube for its interpretation of titles and description.
I suspect also a strong connection between Cloud Video Intelligence for its speech recognition.
Natural language detects sentence syntax (verbs, nouns), sentiment (positive/negative) and topic.
It even detects entities and relevance to other similar topics and websites.
If you want to take Natural Language for a spin, go to https://cloud.google.com/natural-language/ and scroll to section: “Natural Language API-demo”.
YouTube’s official algorithms
YouTube searchThe search algorithm is probably the first algorithm you will think of when it comes to YouTube.
You type in a search query and the algorithm determines by looking at your content, which is the most relevant to your query.
To rank, keyword research is important.
As a small channel: find a topic with high search volume and not much competition.
Do not target a topic that has 100 videos already from big channels, target a topic with just a couple of videos, preferably from smaller channels.
Suggested videosThe suggested videos are the videos on the right side of your YouTube watch page (on the desktop).
This is an algorithm that suggests videos based on:
personal prior activity
topic (related to the current video)
channel authority (gathered watch time minutes of the lifetime of a channel)
Home screenThe home screen is the first screen you see when you visit YouTube.
This is a mix of videos you might be interested in, videos of your subscriptions and videos watched by similar viewers.
NotificationsYouTube has a notification algorithm to notify viewers of a new video on their favorite channels.
This can be the channels you’re subscribed to, but also channels you watch a lot (but are not subscribed to).
I wrote a dedicated article on YouTube notifications for creators.
It goes into detail about the different kinds of notifications, how you can see how many viewers enabled notifications and how to use notifications to your advantage.
TrendingThe trending algorithm displays videos that are popular on YouTube in the country, based on videos that have fast growing views.
Subscriptions tabThis algorithm displays videos of channels that a viewer is subscribed to.
The algorithm makes a selection, so not all videos of all channels are displayed.
The videos that are displayed can be the latest video of a channel, but it doesn’t necessarily have to be.
The algorithm makes a selection of videos to show.
I am sure that there are much more algorithms at work, than only the ones that YouTube mentions in their “common practices”.
Just to mention a few, so you get a whiff of what is going on behind the scenes of YouTube:
- The speech recognition algorithm.
This algorithm translates speech to written text for closed captions (subtitles).
This algorithm gets a lot wrong, but I’ve seen it become better over time.
- Community Guidelines and Ad friendly Guidelines algorithm
An algorithm that checks against violations of the Community Guidelines, like hate speech or discrimination for example.
This algorithm checks all the content: the closed captions, the title, thumbnail (image recognition is another sub algorithm), description etc.
- Filtering comments algorithm
The algorithm that looks at comments and determines if they are spam.
YouTube takes this a step further and can determine if a comment has a positive or negative character and filter on that.
- Video format algorithm
The algorithm that is responsible for converting your video file format into YouTube’s own format with the best possible quality with the highest possible compression rate to save disk space.
- ContentID algorithm
Finding matches in your content that are in YouTube’s copyright protected material library (audio and video).
When you are serious about YouTube and want to take it to the next level, take a look at my video training program: Viral Strategy.
The program takes you step-by-step through the process of getting views, subscribers and going viral.
For new creators I included a module that guides you step-by-step through the process of starting, creating and setting up a YouTube channel.
Share this articleIf you took value from this article, please share it on social media.