What content dominates on YouTube?

YouTube is the most popular video hosting platform on the internet. Billions of people watch billions of hours of content. Tens of millions of users upload content to it every year. These statistics are fairly well known and understood by the wider creator community.

YouTube however, doesn’t share many details about its platform and most best practices are based on anecdotal or past experiences.

Because YouTube is one of the platforms that we, at Pex, actively monitor, we have decided to share some never before published data to help creators across the globe understand this platform better.

Disclaimer on the data

In order to better understand the information shared below, you should know what Pex does. Pex is a search engine, not dissimilar from Google. We index selected platforms, extract audio and video content, fingerprint it and search through it. In addition, we collect and update all surrounding metadata on a regular basis. These include views, comments, likes, dislikes and any other useful information. We have no direct relationships with any of the platforms. We have no visibility into private and unlisted content, nor have we any information about the consumers of the content.

That said, we monitor all visible content on all of the supported platforms. We don’t estimate, extrapolate or in any way tamper with the data we extract from these platforms.

All the data in this article is based on performance of all publicly available videos as of Dec 31st, 2018.

By Dec 31st, 2018, YouTube had over…

  • 5.2 billion videos uploaded
  • 1 billion hours of content
  • 29 trillion views
  • 250 billion likes + dislikes
  • 33 billion comments
Hours of content uploaded every minute

One of YouTube’s favorite metrics to share is how much content is being uploaded. In March 2019, Susan Wojcicki, the CEO of YouTube, publicly shared that more than 500 hours of content is being uploaded to YouTube every minute.

The actual number is even higher. Just within 2018 more than 621 of hours of content were uploaded every minute, or more than 10 hours of content every second.

* July 2010: Video length max increased from 10 to 15 minutes for all users | September 2011: Verified users can upload long-form videos

The length of videos is increasing, driven mostly by long form content, primarily gaming videos and live streaming.

Unique uploads within the year

Users uploaded more than 1.3 billion videos in 2018. As you can see from the graph above, the growth is slowing down.

Unique account that uploaded at least one video within the year

Once again, when we look at the number of users uploading content, the growth is slowing down. This is not caused by competition luring users away, but rather by the fact that YouTube has peaked within the connected population. Of the 4 billion people that now own a smartphone, 800 million are blocked from accessing YouTube in China and some large parts of the World still don’t have fast enough connections to consume video content.

This peak also explains why YouTube spends so many resources on keeping users more active within the platform.

Average uploads per unique account per year

And it’s working. You can see in the graph above that from 2010 and 2016 YouTube focused on a land grab, getting as many users to participate as it could get its hands on. For the last two years however, they are focusing on improving the engagement with their creators. On average, a user uploads 13 videos every year. Almost 3 times more than a decade ago.

All the above data is very generic to the platform itself. However, YouTube breaks content into categories. These categories are predefined by YouTube and selected by users at the upload time. People & blogs is the default category. That means that if a user doesn’t change the category, it will be automatically attributed to it. A full list of all categories can be found here.

Distribution of content by category

The graph above visualizes the distribution of newly uploaded content to the most popular categories. Gaming is by far the fastest growing. In fact, in absolute numbers, YouTube has orders of magnitude more gaming videos than Twitch. That doesn’t make them more popular by the viewers, nor does it makes them good revenue generators.

Break down of each category

In fact, if you look at the table above, you will see that Music, not Gaming, is the most profitable category. This is caused by multiple reasons. Gaming content is by far the longest, which requires YouTube to spend more money on hosting it. Music, on the other hand, is the shortest of all categories, but in turn generates the most views per average video.

Absolute numbers per category in 2018

Even though the Music category received “only” 20% of all views in 2018, it also represented only 5% of all content on the platform. Music and Entertainment are the only two categories that disproportionally deliver high returns (views) on investment (amount of content = hosting & distribution cost).

Also the two most popular categories are also the least “native” to YouTube and YouTube has to license their content.

If you are curious, here you can find list of top 10 videos per each category as of Dec 31st, 2018.

Distribution of videos based on the views

Forget the Pareto principe (80/20 rule). YouTube’s distribution is significantly worse. Only 0.64% of all videos ever reach more than 100,000 views.

Why does it matter?

Distribution of views as % of total views on the platform

Because these 0.64% represent an incredible 81.6% of all views on the platform. You read it right. Should YouTube remove 99.36% of all underperforming videos, they would save an astounding amount of money and still retain most of the revenue (especially considering that most of the underperforming videos are on channels that don’t meet monetization criteria).

Distribution of views per category

Music is the only category that consistently attracts hundreds of millions of users to watch the same videos over and over. The first video that ever broke 1B view mark was a music video. The vast majority of videos with over 1B views are music videos.

Not all content is equal.

How big is music on YouTube?

Several weeks ago, Lyor Cohen, YouTube’s music ambassador, published a blog post on the role of YouTube in the music industry. He pointed out that “there’s still a disconnect between YouTube and the rest of the industry” and outlined 5 factors that, in his opinion, are responsible for the current state of affairs.

Just a day later, the CEO of RIAA Cary Sherman, wrote a reply that questioned many of Cohen’s statements, and backed some of the counterarguments with data.

No matter who you side with, we all can agree that the lack of transparency is not helping the debate. So instead of picking sides, we will try to answer a question: How big is music on YouTube?

But before we do so, you need to understand where the data is coming from. Pex is a search engine not dissimilar to Google or Bing, with a primary focus on video and music.

How Pex works

At its core, Pex crawls the web to identify audio-visual content. When it does so, it fingerprints the multimedia files and also extracts the surrounding metadata, which is then constantly updated (on average every couple of hours). That is why we have clear visibility, not only into the performance of selected content, but also into whole platforms, including YouTube. At the time of writing, Pex has indexed over 6.1B videos and songs across 20+ sites, including YouTube, Facebook, Instagram, Soundcloud, Vk, Youku/Tudou, Twitch, and more.

So, how big is music on YouTube?

One way to answer this question is to count all of the videos that are uploaded to Youtube, and that are registered in the category of Music. However, this approach is flawed, as the categories are self reported by the uploader. Users can choose one of 15 or so categories, to best describe their content, with People & Blogs being the default selection.

You can find the spreadsheet here

In the table above, you can see the breakdown of all the videos uploaded to YouTube by August 22, 2017, sorted by views. Music is by far the most viewed category with an overall traffic of over 27%.

A better way to answer the question above is to look at how many videos contain any music, regardless of the user’s categorization of the video. Thankfully we can do so, because we run a classifier on every single ingested video, which annotates the audio track with labels like “music”, “human speech”, and more.

Based on our calculations, more than 84% of videos contain at least 10 seconds of music.

These results don’t imply the ownership of the music, nor its legal status. At the moment, it’s way too complicated to answer this question.

What we can try to answer instead is: How much of the content containing music is being monetized on YouTube? As Mr. Cohen stated in his article “as of 2016, 99.5 percent of music claims on YouTube are matched automatically by Content ID and are either removed or monetized.

However our data shows that almost 65% of these videos are not claimed, and thus generate no revenue.

What the data doesn’t show is the correctness of the claims. It’s not easy to tell, if the videos are being claimed by the true rightholders, or by users who may be claiming ownership improperly.

Claimed vs not claimed videos over time

If we plot the data over time, the trend seems to be getting worse, not better. Here is a graph showing the difference between “videos not claimed” vs. “claimed videos”, over time.

The difference between claimed and not claimed videos over time

When we were speculating on the reasons why this could be, we thought that maybe Content ID doesn’t have enough reference files to claim more content. To verify this thinking we picked one of the most popular songs on YouTube: “Gangnam Style”.

Of all 891,685 copies we found, 182,220 weren’t claimed (~20%), representing an accumulated 0.5B views.

Perhaps this performance is because the segments containing Gangnam Style were under 20 seconds? With the recent boom of short-form content, this would make sense. However, when we consulted the data, the results paint a very different picture.

The average length of a segment containing the Gangnam Style’s music is 46.6 second for all content that was not claimed. This is roughly three times less than the length of the segments for all claimed content (131.7 seconds), but still long enough for Content ID to identify these videos.

Average length of video uploaded to YouTube over time

In fact, the average duration of all videos uploaded on YouTube surpassed 14 minutes (851 seconds) at the end of 2016, and is currently approaching the 15 minute mark. It seems like truly short content is more commonly uploaded to other platforms, like Giphy or Instagram. If Pex can match segments of videos down to 0.5 seconds, across all platforms, the length of matches and videos on YouTube should be no excuse for Content ID.

Pex Secures $7M in Series Seed Financing

Today, we are thrilled to announce that Pex has raised $7M in Series Seed financing. The round is led by Illuminate Ventures, joined by Universal Music Group, Susa Ventures, Warner Music Group, Blibros Capital Partners, and others.

We began with a mission to build the best search engine for audio-visual content. With big names dominating the search market, we understood early that we would have to use a different approach to achieve our goal. We started with building a scalable system, which indexes audio-visual content across the Web, and delivers results with unchallengeable speed and precision.

The heart of our service is based on a proprietary, in-house built, fingerprinting technology. It is capable of identifying re-used content as short as 0.5s, across our ever-growing archive that currently holds over 6B songs and videos.

On the back of our core technology, we built an online dashboard that allows musicians, filmmakers, marketers, brands, rights holders, and others to understand the true reach of their content through both its virality and popularity. Additionally, we allow all rights holders to monitor and resolve any copyright infringements at scale, with great precision.

For the past 18 months, we’ve been quietly working with the world’s leading media companies, brands and right holders, to ensure that we deliver on the promise of scale, speed and precision. Even though we would love to, we’re still not ready to unveil the service to the general public.

Our focus for the next 18 months is to expand our service to more customers, increase our coverage of monitored platforms to linear television and radio, and to introduce more features, especially around predictive analytics and automation.

If you want to be part of this journey, join us here in sunny Los Angeles.

Why we relocated from San Francisco to Los Angeles

Packing up our old office

Last month, Pex turned 3 years old. But instead of celebrating with a big party, we were busy packing our office and homes, and moving to Los Angeles. I got many questions about what triggered the sudden move, so I decided to share our reasoning publicly.

  1. Better access to highly skilled talent

Pex is a very tech heavy company. These days, we’re operating on 25–30k servers, processing over 32PB of multi-media content every single month, covering more than 4B videos, while adding 50M new ones every single day.

We manage all this with a lean team of just 12 people. As you guessed, not that many people can actually build systems running at this scale. The Bay Area is perhaps the one place that actually has no shortage of these talents. So why move away? Because in SF, we can’t afford them.

Engineers suitable for the role are being fought over and we’re almost always on the losing side. Most companies solve this problem by hiring remote talent, but this just doesn’t work for us. I’ve tried it for 7 years, across multiple companies. We always ended up spending more time on clearing up misunderstandings than on working to solve our customers’ needs.

Being in SF required fighting over talent, not only with the biggest tech companies in the world, but also with thousands of well-funded startups. And of course we came up short in this equation.

LA isn’t yet perceived to be a very high tech city. But for decades it’s actually been the epicenter of the space industry. LA (and surrounding cities) are also home to many popular media tech startups — Snapchat, Oculus and others. For us, this means access to people with complementary skills.

To make sure we can attract, compete and retain talent, we also decided to keep our SF salaries. Of course, higher salaries won’t always guarantee better candidates, but they gives us a fighting chance. We’re already seeing the benefits. In the last couple of weeks, we received resumes from candidates that we would never get a chance to talk to in the Bay Area.

2. A cheaper (and better?) life

San Francisco and surrounding cities are incredibly expensive. There is no point in writing any more about this here, just go read the news.

Even though I love San Francisco and will miss it greatly, I have lost hope that this situation will get better anytime soon. I feel responsible for my colleagues, especially those that we moved from the other side of the world. I don’t want them to have to live with five roommates and be in constant financial stress.

LA allows us to do better. Even though some areas are getting as ridiculously expensive as SF, the rest of the city is far from it. Every single person at the company is doing better in terms of either housing or personal savings. By relieving the financial stress, they can focus more on the job at hand and live happier lives.

3. Being closer to our clients

Our primary customers are content creators, rights holders and brands. A majority of those are located in LA. By being near them, we can be more hands on with their teams, we can attend sales meetings in person and we can gradually educate the market. Being here helps us to understand our clients better. We’re building the service for them, so being able to meet them on their turf and listen carefully to their feedback helps us to do better.

What we didn’t anticipate though was the reaction of the customers. Once we settled in, we started letting people know that we have moved. We especially focused on companies that, for one reason or another, weren’t interested working with us before. We didn’t expect any miracles. But as they learned about our relocation, they warmed up. When I asked why, I was told that by moving to LA, we’ve shown dedication to the business.

Even after few good signs, it’s too soon to judge if our expectations will be met in the long run. I will be sure to report back on our progress once we get settled in and have time to dig around a little bit more.

Oh and yes, we’re hiring.