TikTok knows me better than I know myself. It can sense when I want to watch a dog do a weird little scream (always) or how often I’d like to see couples working out together (never). Fellow TikTok obsessives often attribute the app’s eerie precision to its all-seeing eye, its own enigmatic Big Brother: the Algorithm. But how exactly do TikTok and its algorithm know what we want to watch? A new report from The Wall Street Journal looked at how TikTok’s algorithm works, providing some insight into why your For You page feels so hyperspecific to you.
To conduct their investigation, the Journal created over 100 automated accounts (i.e., bots) with a handful of identifiers: an age, a location, and a couple of specific interests. Each bot watched hundreds of thousands of videos, rewatching or hovering on TikToks that fit into its preferred content. (Which, honestly, sounds pretty similar to how I, a human, behave on the app.) TikTok learned the interests of most of the Journal’s bots in less than two hours, some in less than 40 minutes.
While TikTok prioritizes content that’s similar to what you like, share, and follow, the Journal found that one factor appears to usurp them all: watch time. In other words, TikTok knows which videos you watch in full, what you rewatch, what you scroll past after a few seconds, and curates its recommendations accordingly.
Given its dubious history with users’ privacy, a spokesperson for TikTok who spoke with the Journal was adamant that the app doesn’t listen through your microphone or read text messages to serve personalized videos. This echoes a June 2020 blog post from TikTok — posted around the time the Trump administration started talking about a TikTok ban — that explains how the app recommends videos to users. According to the blog post, information including content you like, comments you post, hashtags on videos you watch, your device type, and your location settings are all weighted to feed your For You page with TikToks the algorithm thinks you’ll enjoy or, at least, watch.
While the blog post doesn’t explicitly say how much of a factor watch time is, it does say that “a strong indicator of interest, such as whether a user finishes watching a longer video from beginning to end, would receive greater weight than a weak indicator, such as whether the video’s viewer and creator are both in the same country.” Like the Journal report found, what you watch heavily informs what TikTok’s algorithm feeds you.
“The algorithm is able to find the piece of content you’re vulnerable to,” Guillaume Chaslot, founder of Algotransparency, told the Journal. “It doesn’t mean you really like it, that it’s the content you enjoy the most,” Chaslot noted. “It’s just the content that’s the most likely to make you stay on the platform.” In other words, TikTok doesn’t care whether you’re watching out of enjoyment or spite — so long as you keep watching.