Why is TikTok the reigning champ of discovery?

How TikTok’s video format leads to a better algorithm

When Spotify launched Discover Weekly a few years ago, I was obsessed with it. It’s was (and is) an incredible way to find new music. It also wildly increased the value of Spotify from being a glorified online iPod to a music platform built around you.

For listeners of Rabbit Hole, we all know that Youtube also plays a big role in content discovery. Their mission to hit a daily 1 billion hours viewed led to a recommendation engine that has drastically shifted viewing habits.

Still, between Youtube, Spotify, and TikTok, TikTok is the clear winner in content discovery. By the nature of the platform, TikTok has to serve you 4–10x more pieces of individual content than Youtube and Spotify. The fact that they are able to do that successfully is nothing short of incredible.

TikTok’s platform is unique because:

• all videos are capped at 3 minutes (and most are under 60 seconds)
• users engage with more individual pieces of content than on other platforms
• searching for TikToks is not the ideal way to consume content
• posting is easy

The Unique Information TikTok’s Format Provides

TikTok’s 3 minute limit on videos provides a potential problem. TikTok has to serve its users a lot more content than Spotify, Youtube, or Netflix. 30 minutes on Spotify might look like 10 songs, but on TikTok it can look like 40–60 videos.

Given the above, users inherently provide TikTok with a lot more data. Though I imagine TikTok’s algorithm uses numerous inputs to track engagement, the below are a few points that could be useful and exist solely because of the app’s content format:

For any individual TikTok, the following stats could be used:

• % of time engaged (1 view is 100%, 3 views is 300%, 1/10 of a view is 10%)
• Seconds engaged (note that this is different than a % of time watched)
• Times the user has engaged with the sound (Is this an original sound clip or is it a viral sound, much like the one used in the adult swim trend from earlier this year?)
• Filters and Effects (is the user drawn to videos with certain filters?)

Though there are a few other important inputs one can use to build a good algorithm, the above four lend themselves to surfacing incredibly relevant and trendy content.

For simplicity, let’s tear down the first point as it relates to a competing platform. The average length of a Youtube video is 11 minutes. This means that finishing a Youtube video likely means that I enjoyed the content, but even if I loved the video, I probably won’t rewatch it immediately. On TikTok, I find myself rewatching my favorite videos 2–4 times in a row. The short-form nature of TikToks lend the algorithm an extra data point to determine if a watched video was “liked,” “loved,” or something in between.

The User’s Perspective

Regardless of what information the algorithm gathers, why are TikTok’s users so willing to provide it? For new users, your first few minutes (and hours) teach TikTok what you like. That means you’ll almost certainly see some things that don’t resonate with you — shouldn’t that result in people getting frustrated and leaving?

The Time to Buy In Problem

Once again, TikTok benefits from its three minute video cap. It differentiates TikTok by solving what other platforms cannot figure out: the Time to Buy In Problem.

The “Time to Buy In Problem” is an equation* that suggests the amount of time it takes a user to “buy in” to a piece of content is a sum of minutes to discovery and minutes to meaningful content.

Here it is, more simply put.

Time to Buy In = minutes to discover content + minutes to meaningful content

Pretend it’s Thursday night and you’re watching something on Netflix. Your time to buy in for any given movie might be:

Netflix Time to Buy In = ten browsing minutes + five minutes watching

Once a user has “bought in” to a piece of content, they are far more likely to finish it. I personally see that with longer pieces of content, my Time to Buy In is often longer. It makes sense! Two hour movies require a bit more setup than a 60 second video. Still, that doesn’t make that first 15 minutes fun.

TikTok’s equation looks more like this:

TikTok Time to Buy In = 0 browsing minutes + 0-3 viewing minutes

Most users do not search for videos on TikTok. Unlike Youtube or Spotify, 95% of my time is engaged watching content that is chosen by the platform itself. That means that TikTok has completely removed the need for discovering content. They do that for me.

Admittedly, that doesn’t mean everything I watch will be good. My “time to buy in” may be the duration of the video — I may never truly “buy-in,” but because most of the TikToks I watch are so short, it’s really not a high price to pay. This suggests an answers the above question: if the time to buy in is small enough, users will be more forgiving of an individual piece of bad content. I can wade through a few minutes of bad videos to get to the good ones. On Netflix, I’m investing anywhere from 20–120 minutes for a single piece of content. It better be good! On TikTok, it’s a fraction of that.

A Brief Conclusion

TikTok has been able to invent a content form that drives discoverability. For TikTok, a disliked video means a better algorithm, but for Netflix, it can mean a cancelled subscription. It’s an interesting case study on how the form of the content completely melds the platform that hosts it.