YouTube statistics highlight one of the problems with the creator economy

YouTube is the new Hollywood right? But is a secretive platform driven by an all-powerful and clandestine algorithm really the best bet to be hanging a whole new economy on?

YouTube is now 20 years old. In those two decades it has gone from the simple tagline of ‘Broadcast yourself’ to claiming it is ‘the new television’ and ‘the epicentre of culture’. As the traditional industry struggles with declining audiences and shifting economic sands — Panavision Hollywood is reportedly shuttering its offices next month, and the VFX industry is seemingly in ongoing triage — the idea is that the creator economy continues to expand at breakneck speed and cushions the impact.

Vendors are chasing it with lower price versions of pro-level products, trade shows are actively courting a new audience of YouTubers and vloggers that once they would have been extremely sniffy about, and estimates about its growth are approaching off the charts. Analyst Evan Shapiro reckons that it has grown 25% over the past 5 years (as opposed to 3% for traditional media) and is on course to represent a massive $6 trillion economy in 2030. 

All of this is true. But there is a huge sting in this particular tale in that so much of this growth and so much of this future livelihood for so many people will be dependent on the black box workings of one platform. Sure, from Substack to Patreon there are a lot of ways for creators to reach audiences, and, of course, there are multiple competitors in the video arena too, but YouTube is big enough to cast its own distortion field. It’s rejigged its TV app’s interface to be more TV like, added features such as season, and the net result is that viewers are watching, on average, over 1 billion hours of YouTube content on TVs daily. TV is now the primary device for YouTube viewing in the US.

Gaming YouTube’s algorithm is, of course, a black ops science as old as the platform itself. And there has been plenty written about the negative effects of its recommendation system, including worrying radicalisation pathways and similar issues. Social media gets a huge amount of flack for, say, serving beauty ads to teenage girls who’ve just uploaded and then deleted a selfie, as Sarah Wynn-Williams’ Careless People details amongst many other mundanely horrifying things. And rightly so. But the sense is that YouTube gets less attention here, and perhaps its positioning itself as the new TV is an attempt to make people think less about that particular aspect of its business model.

But for such a critical piece of new economy infrastructure you can’t help feeling that it would be good to know a little bit more about its workings. Which is why some researchers spent a couple of years ‘drunk dialling’ it to find out what it has on its servers.

A peak under the hood

You can read the full details of the research at BBC Future, but the long and the short of it is that a team at the Initiative for Digital Public Infrastructure at the University of Massachusetts at Amherst in the US worked out a way to bypass the algorithm’s working and get hold of a genuine random sample of YouTube videos. It did this by creating a program that randomly generates a string of 11 characters and sees if there is a corresponding YouTube video identifier that matches it. If there is, it downloads it. As the Initiative’s director Ethan Zuckerman puts it, you can think of it like a pesky teenager punching in random numbers for prank calls after dipping into his parent's liquor cabinet.

The only thing here is that the amount of processing it has to do is infinitely greater than that involved in randomly selecting a 10 digit phone number. YouTube’s combination of letters and numbers leads to 18.6 quintillion potential combinations and the scraper the team built in the end tried an average of 1.87 billion bad guesses for every one video it found.

The result was that after guessing at 18 trillion potential URLs it ended up with an initial batch for analysis of 10,016 random videos. That’s a decent enough sample size to extrapolate out to the rest of the vast catalogue, and once it crunched the numbers on them it came up with some surprising data points.

Here are the highlights:

• The majority of YouTube videos get under 500 views
• In fact, the average number of YouTube views is 41
• The median length is 64 seconds
• English language content makes up 28%, followed by Hindi and Spanish (see below)
• 4% of videos have never been watched even once
• Only 0.21% feature any form of monetisation
• Under 4% feature any form of call to action
• Only 38% have undergone any form of editing
• Only 18% were judged to have high quality audio

Hidden meanings

The point is that all of this had to be prised out of YouTube. The amount of information it freely shares is increasingly negligible as it pretty much tries to hide in plain sight and avoid the regulatory heat that other algorithmically driven social networks have had to navigate. It wants to be the TV for the next generation, but it is also a dumping ground for pretty much all and any video shot around the world, has been proven to have some quite negative impacts (it is a known conduit for radicalisation, for instance), and can alter livelihoods and people’s lives with a single tweak in the calculations regarding what it recommends and who it recommends it to.

The saga of Facebook and the publishing industry offers a cautionary example of what can happen. Essentially Facebook lured them in, then it decided to deprioritise their content, then it insisted they would have to pay to have the same reach as they had before, by which time they were dependent on the site for traffic. Then it decided it just didn’t need them anymore and moved on to something else. Thanks, Zuck.

But as Hollywood and television retracts, so more and more of us will be working in the creator economy that is exposed by default, and frankly we deserve better than that. The question is, with YouTube as it currently is, whether we will get it.