See more in Data

Why YouTube's Algorithm Constantly Recommended Watching Gangnam Style

4 min read

I'm Jeremy Carson, and this is everything I wish I knew about the advertising and creative industry when I got started. And everything I'm discovering as a Creative Director today.
LinkedIn Instagram TikTok 

YouTube broke out of being a video hosting service and into a content delivery machine with one tool: their recommendation engine.

It creates the "YouTube Vortex" we all get stuck in. You know the one: where you started watching TED Talks about cold fusion and eventually found yourself engrossed in fail videos of people making slime and destroying their kitchens.

It's a fantastic algorithm, earning them billions because of how long it keeps people on the site. But it's constantly evolving, with machine learning and tweaks to the code. One of these tweaks happened a few years back, when something called "The Gangnam Style Effect" took hold of YouTube.

There's this claim that algorithmic recommendations create echo chambers. If you're constantly told, "Hey, here's another thing like this thing you like!" then you're not really exposed to challenging ideas or concepts. Netflix, Spotify, Pandora, YouTube, and tons of others are desperately trying to fix this issue, while at the same time, keeping their customers happy.

And with the rise of data in advertising, modern marketing is at the same risk. But, let's take a look at how it affected the biggest video-viewing site on the planet.

The Gangnam Style Effect

See, the way the YouTube algorithm worked was, after every video, it would autoplay another one that was broadly related, but had more views. Now, theoretically, if you left YouTube playing long enough, every video, down the road, would end up leading to Gangnam Style (that K-Pop music video from Psy), since it's the most-viewed video on the site.

YouTube realized that that a "show me a more popular video" strategy wasn't the best use of their data. It made it really hard for content creators to develop a following. Plus, it created a homogenized viewership, one that expected that, ultimately, everyone would want to see the same things. Which, aside from Gangnam Style and Avengers: Endgame, is rarely true.

So, they changed their algorithm. Instead of showing you a more popular video related to yours, it would show you an even more niche video related to it. Their claim was that this would give smaller content creators an opportunity to shine.

What they didn't expect was it created a whole other problem: echo chambers. Deep niche and fringe ones.

Data Creating Echo Chambers

It makes perfect sense: the more you cater to someone's interests, the more they'll consume. But what YouTube found was that by recommending niche related content, they simply polarized their echo chambers even more.

By knowing what people expect, we can challenge it.

We see it in politics and religion all the time. You surround yourself with people who think exactly like you, and you'll go into a tailspin of agreement. And the more fringe your opinions are, the easier it is to galvanize them. And YouTube's enabling it all, through their data strategy.

But all they wanted to do was show people new videos, so that they could grow their platform.

Advertising and Data

It's the same concern in advertising: showing people what they want is good in the short term, but bad for growth. People have their opinions about brands, and to change that opinion, you need to go against the grain.

Which is why we need to use data in a very careful way in advertising. Now, performance marketing is somewhat immune to this. Their whole purpose is to show people what they want to see, so they can get them to convert to customers. They can use data to interpret customer behaviors and go with the flow.

But on the brand side of things, we have to be careful. Data can't simply be a playbook for what to say. Like anything, it's simply a tool, one that we need to use in the right way. A hammer can be used to smash a nail into wood, but it can also pull it away. And the same goes for data.

Data can be used to know what people expect to do, or see, or hear. And then we can use that data-driven insight to push against that behavior. By knowing what people expect, we can challenge it.

Pushing Against the Data

While YouTube's constantly adjusting its algorithm to drive growth rather than echo chambers, other services have started to figure out ways around it.

Spotify's "Daily Mix" has been a major hit among its users. People love that it "just gets them." But what they probably don't realize is the mix of songs you love is probably hiding some songs you may not know you'd love. Spotify can use the "Daily Mix" to grow their listeners' music-listening habits by sneaking in the songs the data says they may or may not like.

Same goes for Amazon. You know that row of products at the bottom; the products that "people also viewed"? Well, that's the perfect place to introduce a tangentially-related product, isn't it???

There's a theme: use the data to understand what people might like, but present it alongside things the data says they will like. It's a little sugar with the medicine.

Use the Data, Don't Let It Use You

Don't be a slave to data. Don't let it dictate what you should do.

Bring insight to it. Bring creativity to it. Bring a voice that doesn't echo what everyone thinks they want to hear, but instead say what they need to hear.

Henry Ford is famously quoted as saying, "If I gave people what they wanted, I'd build a faster horse." He knew that the data (people's opinions) said "faster horse," but really meant "they're getting there too slowly."

Human creativity figured out how to interpret that data and create against it. That's why we need to build with our hearts and our heads.

Thanks for reading!

If you enjoyed this, say hello @thejeremycarson. LinkedIn Instagram TikTok