YouTube, new details on algorithms and ranking
Videos are becoming an increasingly central channel in marketing strategies: according to Cisco, within the next two years online will account for over 82% of all Internet traffic of consumers, users watch an average of 16 hours of online videos per week and already 85% of companies use them as a marketing tool. The main reference in this field is still Youtube and so it is very useful to find out how its ranking system works and what factors influence the video recommendation algorithm, thanks to the new details revealed by the members of the team responsible for this work.
YouTube algorithms, we still have little info
The information comes from two new Q&A videos on Youtube’s recommendation algorithm, both reported by Matt Southern on Search Engine Journal, commenting on the questions and answers presented and reminding us that the Youtube machine learning algorithm was only implemented in 2016, so we still have a rudimentary idea of how it works.
At the moment, we know that video recommendations are influenced by factors such as clicks, viewing time, number of “Likes/Dislikes”, comments, freshness and upload rate, already revealed in other insights on SEO for Youtube.
However, we do not know “whether external traffic has an impact on recommended videos”, or “low-performance videos influence the likelihood of future videos being recommended”, or again “what is the impact of other potentially negative factors, such as inactive subscribers or too frequent uploads”, and it is precisely on these points that Q&A focus.
- How YouTube ranks search results?
The first question addressed is perhaps the most “trivial”, the one that everyone would like to do: how does the ranking on Youtube work? Obviously we are not told who knows what secrets, but the team of the video platform uses terms that lead us immediately to the classification system of Google.
In both cases, in fact, the goal is to show users the most relevant results for their queries, and then Youtube classifies videos according to a series of factors, the most important of which are precisely the relevance and performance.
Relevance is the way in which the title, description and content of a video match the query of the user; performance is related to the videos that users have chosen to watch after running similar queries.
Moreover, the Youtube algorithm also considers engagement metrics as the duration and amount of videos users choose to watch.
To further clarify, then, Youtube search results are not a list of the most viewed videos for a given query, but a reasoned list of videos that are more relevant and that are more likely for the user to watch.
- Non-performing videos do not affect the channel’s performance
We then move on to analyze in more detail the system of “recommendation”, starting from a question on the impact of a video with poor performance and lower than the others on the channel and on the performance of other videos.
The response of the Youtube team is clear: the algorithm “does not make assessments on a channel as a whole based on the performance of some videos”, but only cares about the ways in which “people respond to a certain video when they decide whether to recommend it to others”.
Therefore, the Youtube recommendation algorithm will always try to “follow the audience“: if a video is attracting an audience, it will be shown in the user recommendations regardless of the performance of previous videos of the channel. It is “normal for video performance to fluctuate in terms of views and other metrics,” but Youtube is careful not to base all its advice on such metrics.
It follows that channels should not worry about something like “an algorithmic demotion based on a decrease in visualizations”.
- Impact of the change of title and thumbnails
Speaking of videos that do not have sufficient performance, there are simple techniques that can help in reversing the course.
In general, making changes to a video is only recommended when it has a lower click percentage and – at the same time – receives fewer views and impressions than usual. In such cases, Youtube strongly recommends changing the title or appearance of the video preview thumbnail, because this can be an effective way to get more views.
This intervention shows the video with a different appearance for the viewers and this could change the way people interact with it when they find it among the recommended. It should be clarified, however, that the Youtube algorithm responds to the change in the user’s behavior towards the changes, not at the time of changing the title or the thumbnail (which in itself does not intrinsically activate Youtube to increase the impressions of a video).
- How important view time is
A user question specifically asks whether “a few hours of watch-time is needed before a video is recommended by the Youtube algorithm”, and the answer is clear: “there is no particular threshold that a video needs to meet before it starts to be recommended”.
Moreover, channels may notice that some of their videos are gaining momentum months after the release of a video “because it is normal for users to show interest in older videos“: this may be due to the fact that a given topic is becoming increasingly popular or the behavior of new viewers of a channel, who go back and retrieve previous videos.
Most users do not watch videos in chronological order, starting from the most recent, nor decide what to watch based on when it was published: so the home page of a user will often contain videos published weeks, months or even years ago.
- There is no daily loading limit
It continues with the answer to another frequent doubt: “Is there a point where the number of videos per day / week on each channel is so high that the algorithm is overwhelmed and the videos slip away?”.
Also in this case we can be reassured, because there is no limit to the number of videos “that can be recommended to a certain viewer from a channel in a single day”. Channels can upload as much as they want and the number of views each video receives only depends on the preferences of the viewer.
Youtube’s recommendation system will continue to recommend videos as long as viewers continue to watch them: so if a channel “is uploading more videos than usual and each video receives a progressively lower number of views, could be a sign that the audience is running low“.
While there is no limit to the number of videos that Youtube will recommend from a channel in a single day, “there is a limit to the number of notifications that will be sent” for each new upload and “Youtube only allows 3 notifications per channel in a 24-hour period”.
- Inactive subscribers does not affect the channel’s performance
There are many questions and doubts regarding the impact of old or inactive subscribers to the channel that, in the fears of the channel owner, could lead to a lower CTR and consequently adversely affect the chances of the video being recommended.
The answers of the platform team try to clear the field from these concerns and to reassure: the Youtube recommendation algorithm does not focus on the subscription feed as the main signal, because it is based on the performance of a video in the context in which it is shown.
Home page placement, for instance, is based on video performance when shown on other users’ home pages.
Moreover: the algorithm is able to understand which viewers have not watched the content of a channel for a long time and will avoid showing the content of that channel to inactive subscribers. And so inactive subscribers are not something that channel owners should be worried about.
There is then a basic “mistake” in the approach to the Youtube recommendation system, which “does not transmit videos to anyone” but inserts the videos and ranks them for users based on what they are most likely to watch, regardless of whether they are published by the channels to which the user subscribes.
And so, being subscribed to that channel is just one of many signals used to classify videos for users: during the tests, Youtube has indeed found that giving priority to the content of the channels to which a user subscribes drastically reduces the number of videos that users watch and the frequency with which they return to the platform.
Reassurances come consequently also on the more general management of inactive subscribers, an issue that worries especially the “older” channels and that pushes many to consider the opportunity to open a new channel on which to reload videos to make them more “acceptable” for the algorithm.
In fact, Youtube team leaders explain that “inactive subscribers are not a factor influencing Youtube’s recommendation algorithm”, provided that (as stated in the first question) the channel continues to attract audiences.
In this case, “if it attracts an audience a channel with inactive subscribers can still see its next video shown in the advice section”, while “creating a new channel and reloading the same videos will not help to show those videos to more people”. Youtube remembers viewers’ preferences and “there is little chance of reaching those inactive subscribers with a new channel”.
Creators should only start a new channel if they decide to go in a different direction with their content.
- Videos in multiple languages on the same channel can be confusing
The next issue addresses the possible impact of uploading videos in two different languages on the same channel on how videos are recommended by Youtube: This situation can cause confusion for viewers and therefore Youtube suggests to create separate channels for each language.
However, if the channel is targeted specifically at an audience that speaks multiple languages, it makes sense to keep all the content on the same channel.
- The impact of external traffic
The latest focus is on the value of external traffic, which “is definitely a factor influencing the Youtube recommendation algorithm” but whose impact does not go beyond this element.
Specifically,”external traffic can help to make a video appear in the recommended section, but once there, it must work well with the viewers“.
The long-term success of a video depends on the reaction of the viewers who clicked on it and found it among the related ones.
Similarly, you should not worry too much even if “the average length of the display decreases when a video receives a significant amount of external traffic”, because it seems to be a common circumstance that, However, it has no impact on the long-term success of a video.
Once again, Youtube’s recommendation team managers explain that the algorithm cares more about the engagement of viewers with a video on which they clicked between those recommended to them than other parameters, and “does not deal with what viewers do after clicking a video from a website or an external app“.
