The algorithms have some behaviorial issues. All over social media, people are complaining about the content recommendation engines have been spinning up recently. It’s too random. It’s “terrible…unusable…like junk mail.” It’s even, at times, oddly biblical. This is a dramatic reversal from an earlier cultural moment, when Instagram’s and TikTok’s algorithms unnervingly seemed to know their users better than those users knew themselves.
Some of the current dissent is coming from everyday Instagrammers who have noticed unwanted changes encroaching on their feeds. But this week also saw a Category 5 Kardashian-Jenner hurricane blow into town when Kylie Jenner and Kim Kardashian both reposted a protest meme reading “Make Instagram Instagram again.” The backlash forced Instagram to reverse some of its proposed changes to the app—the company now is phasing out a full-screen test version of its app and reducing the number of recommended posts that users see.
The last time Kylie Jenner complained about a social media site, Snapchat lost $1.3B, so yeah Instagram has a problem. pic.twitter.com/ugipc9abb6— Frank Pallotta (@frankpallotta) July 25, 2022
But what, besides blowing up Instagram CEO Adam Mosseri’s Twitter mentions, can the average user do about an algorithm gone awry? I faced that question earlier this year when my TikTok For You page became inundated with adult entertainers offering sexually explicit pointers. Those unwanted posts turned up after I’d watched a handful of (mostly safe for work) videos from the account of Quinn, an audio porn company whose founder I was writing about. So what would it take to rein in my rogue feed and return it to serving me the puppy videos and beauty tutorials I preferred?
To learn how to retrain my recommendations on TikTok, Instagram, YouTube and other social sites, I spoke with a few algorithm coaching experts. As social platforms double down on their commitment to recommended content, they told me, the frequency of algorithms going rogue is likely to increase. But there are still ways to mold a recommendation engine in one’s own image. Here’s how to start.