How to retrain your TikTok algorithm

TikTok is the only good social media app. At least, it used to be. From 2019 to 2021, the algorithm was magical. It had a way of sending videos that you didn’t know you were interested in until you saw them. For me, that meant synchronized dancing, vegan recipes, and elaborate songs created from internet dramas. Then came the libel trial of Amber Heard and Johnny Depp.

For weeks this summer, my TikTok feed was hijacked by bizarre videos praising Johnny Depp and offering violent, hateful comments about his ex-wife. No matter how many times he clicked “I’m not interested” or walked past without looking, more would appear. Whether it came from actual fans, opportunists, or bots, the pro-Depp content was relentless.

When the case ended, so did the videos. But my algorithm has never fully recovered. Instead of jokes, I landed on a side of TikTok that’s full of protein powder recommendations and gory surgery videos. It’s like a ’90s kids magazine come to life.

Holding on to memories of the good old days, I’ve spent the last few weeks trying to retrain my algorithm. It turns out that this is not easy. It’s not just a matter of liking the videos you care about. You have to quickly skip the ones you don’t like and save the ones you like.

TikTok’s algorithm is supposed to be particularly insightful because the app collects a lot of data. The company, owned by China’s ByteDance, keeps a close eye on how long a user spends watching a video. It adds that to the information it collects about your location, age, gender, and search history. There is also a lot of content to analyze. Videos can be recorded on smartphones and easily uploaded. They are short, so they can be seen more than once. The fact that all of this data is being collected by a Chinese-owned company is something that has raised frequent security concerns in the US.

However, what keeps over a billion users like me hooked is that TikTok also shows random videos from time to time instead of boring users by showing the same type of content. In the US, TikTok is watched longer than YouTube, according to data taken from Android phone users by analytics company App Annie. Rival social media companies have given the app the highest compliment by trying to incorporate TikTok-like formats into their own platforms. For Instagram, it’s Reels. For YouTube, it’s Shorts.

As I discovered, however, randomness can also be a source of problems. Unlike Twitter or Instagram, which focus attention on the content of the people you choose to follow, TikTok serves strangers. That can leave feeds open to nasty videos. Last year, the company said it would try to improve this and prevent users from falling down rabbit holes of disturbing content. But I suspect that many creators have become more adept at working with the algorithm and moving certain content forward.

The final step I took to improve my TikTok algorithm was to block creators whose videos repeatedly showed me and filter out certain words. It’s laborious but seems to have worked. The disgusting videos have been replaced by remixes of a kid talking about how much he loves candy corn.

Still, TikTok’s glory days of endless pranks and cleverly witty videos seem to be coming to an end, at least on my feed. Once the celebrities and advertisers moved in, the quality of the videos on the app dropped. Even after doing some algorithm repair job, you may need to find a new way to mess around.

Elaine Moore is the deputy editor of Lex of the FT

Leave a Comment