On his personal blog called Remains of the Day, former Amazon and Oculus product executive Eugene Wei wrote Seeing Like an Algorithm, an in-depth analysis of the TikTok algorithm explaining exactly what makes it so special (and performant). In order to help those of you who have been on a mission to try and understand TikTok’s recommendation engine, we rounded-up the key takeaways from Wei’s analysis. Here’s how the design of TikTok helps its algorithm work as well as it does.

Although many link TikTok’s impressive success to its For You page, which isn’t wrong in itself, the video-sharing app relies first and foremost on its algorithm. After all, without the perfect algorithm, even the For You page would collapse. “Understanding how the algorithm achieves its accuracy matters even if you’re not interested in TikTok or the short video space because more and more, companies in all industries will be running up against a competitor whose advantage centres around a machine learning algorithm,” explains Wei.

There, we’ve said it: TikTok’s actual machine learning (ML) recommendation algorithm isn’t out of the ordinary. As Wei explains, “Most experts in the field doubt that TikTok has made some hitherto unknown advance in machine learning recommendations algorithms. In fact, most of them would say that TikTok is likely building off of the same standard approaches to the problem that others are.”

While this statement might throw you off a bit, keep in mind that the effectiveness of a ML algorithm isn’t a function of the algorithm alone but of the algorithm after it is trained on some dataset. Once trained on an enormous volume of data with a massive number of parameters, its output is often mind-blowing.

The TikTok For You page algorithm, trained on its dataset, is remarkably accurate and efficient at matching videos with those who will find them entertaining. It is also just as good at suppressing the distribution of videos to those who won’t find them entertaining.

TikTok needed an algorithm that would excel at recommending short videos to viewers, and when it first launched, “no such massive publicly available training dataset existed.” Even though you could find short videos of memes, kids lip-synching, dancing, cute pets—and on and on—you weren’t able to find comparable data on how the general population felt about such videos. “Outside of Musical.ly’s dataset, which consisted mostly of teen girls in the US lip-synching to each other, such data didn’t exist,” writes Wei.

Knowing that the app faced the problem in which the very types of video its algorithm needed to train on weren’t easy to create without the app’s camera tools and filters, how did TikTok manage to get the most valuable inputs possible for its algorithm?

TikTok’s design offers only one video at a time with a number of indicators as to whether or not the user likes it (length of viewing, re-watches, likes, comments, song choice, video subject, shares). This closed-loop of feedback then inspires and enables the creation and viewing of videos on which its algorithm can be trained. It’s as simple as that: “for its algorithm to become as effective as it has, TikTok became its own source of training data,” clarifies Wei.

Typically, user experience (UX) design is meant to be user-friendly. However, to improve its algorithm, TikTok has made its product a tiny bit less user-friendly, as users having to scroll through multiple pieces of content on apps such as Twitter or Facebook is a more ‘frictionless’ experience than just a single video view like on TikTok. The dominant school of thought in tech has centred around removing friction for users in accomplishing whatever it is they’re trying to do.

When you open the TikTok app, it takes you straight to the For You page and goes right into a video, which fills your entire screen. This is not a scrolling feed. It’s paginated, effectively, explains Wei. The video autoplays almost immediately (and the next few videos are loaded in the background so that they, too, can play quickly when it’s their turn on stage).

This design introduces the user to an immediate question: how do you feel about this short video and only this short video? If you watch it more than once, then you liked something about it. If you shared it with someone, then you must have felt something special there too. If you tapped on the spinning track and watched more videos using that song, this is also an indicator that something in that specific video appealed to you. All the information listed above is what TikTok’s algorithm ‘feeds’ on.

With such clear signals—whether positive or negative—TikTok can quickly understand a user’s preference and serve up more similar content. This in turn creates a tight feedback loop and kicks off the flywheel that continually improves TikTok’s recommendations and data inputs.

But even before gaining all that information, the app’s algorithm looks at the video by itself and classifies it. It knows what the video is about, what filters and songs it uses, if there’s food in it or simply human faces, hands or gestures. The algorithm’s vision AI starts by classifying all this.

So, even before you’ve watched a specific video, TikTok’s algorithm already knows the types of videos you have previously enjoyed, the demographic or psychographic information that is known about you, where you’re watching the video, the type of device you have, and more. “Beyond that, it knows what other users are similar to you.”

After you’ve watched the specific video the algorithm has recommended on the For You page, it can now close all the feedback loops and take every one of your actions on the video and can guess how you, with all your tastes, feel about this video, with all its attributes.

Now, how is this so different from other social media platforms such as Facebook, Twitter and Instagram?

Now, if you compare TikTok’s UI with a traditional social feed that offers an endless scroll of content, you’ll notice that the user inputs are less clear. Instead of serving you one story at a time like TikTok, these apps display multiple items on screen at once. “As you scroll up and past many stories, the algorithm can’t ‘see’ which story your eyes rest on. Even if it could, if the user doesn’t press any of the feedback buttons like the Like button, is their sentiment towards that story positive or negative? The signal of user sentiment isn’t clean,” explains Wei.

In both cases, infinite scrolling feeds are ideal. However, TikTok took it one step further by allowing users to only watch one video at a time, gaining clearer data from this and then feeding it back to its algorithm, making it more competent than ever. “If you click into a text post by someone on Facebook but don’t comment or like the post, how can Facebook judge your sentiment toward that post?” asks Wei.

That’s also why some networks that are built around interest graphs like Reddit have incorporated down voting mechanisms in order to serve users the most interesting content. That means weeding out uninteresting content as much as it does surfacing appealing content. Although TikTok doesn’t have a downvote button, by showing you one video at a time, it can notice your lack of interest depending on how quickly you swipe to the next one or which positive actions you take or don’t take.

Wei states that “Triller may pay some influencers from TikTok to come over and make videos there, Reels might try to draft off of existing Instagram traffic, but what makes TikTok work is the entire positive feedback loop connecting creators, videos, and viewers via the For You page algorithm.” You heard the man, competition will struggle before it gains the same popularity TikTok has.