Online video can be entertaining and educational but also addicting or outrage-inducing. It’s especially easy to fall into a trap on TikTok. That’s because of its heavy reliance on its algorithm to feed you content, rather than primarily showing you videos from creators you follow. When an app relies on your unconscious, automatic actions, such as how long you linger over a video, and ignores what you explicitly tell it you want to watch, it feeds your basest impulses.

But there’s a flip side to it: With a bit of effort, you can train TikTok’s algorithm to work for you rather than to exploit your attention to keep you watching. Here are some ideas on how to do that. These are based on TikTok’s documentation, the occasional study that’s been done, various blog posts describing attempts to reverse engineer the algorithm, and my own experiments with the app. Of course, the specifics of the algorithm might change, but the high-level ideas in this post should still be useful.

Think of the ocean of content out there as made up of three seas:

  1. Stuff you want to watch.
  2. Stuff you didn’t know you wanted to watch.
  3. Stuff you don’t want to watch.

Let’s talk about each of these.

Most of us have varied interests: Each of us has perhaps dozens of different kinds of things we like to watch or are curious about. I’ll bet that although TikTok’s algorithm is good, it hasn’t learned all of your interests, because there are millions of possible interests. If you want to see videos about a specific niche musician, or the town you grew up in and are nostalgic about, those are unlikely to show up on your For You Page. But there’s a simple way to fix that: Search for topics you’d like to see and follow one or two accounts related to each of those. Once TikTok shows you a video from a niche topic and notices that you watched it, the learning will kick in and it will gradually make it a regular part of your diet. That will make your feed more diverse and interesting, and less likely to lead to a rabbit hole dominated by a single topic or a small number of topics.

My favorite part of TikTok is discovering things I didn’t know I wanted to watch. The algorithm is pretty good at this! When it shows me something totally different from what I’ve watched before, I usually have the urge to scroll right past because I want my familiar content. But I’ve learned to stop myself and give new topics a chance. If you want to teach the algorithm that you like something, watch it all the way to the end; it may be even more effective to let the video loop and watch it more than once. On the other hand, according to a Wall Street Journal investigation, tapping “heart” on a video is not effective: The algorithm ignores that action.

Finally, there’s the stuff you don’t want to watch but that keeps coming up anyway. There’s so much trashy stuff on TikTok that in the process of trying new content to see if you like it, TikTok will keep serving you these videos. The obvious way to resist this is to scroll right past it to let the algorithm know you don’t like it, but that’s hard. After all, that’s why low-brow content performs so well on the platform. There’s another way to get it off your For You Page—an option that’s both easier and more effective. Don’t blame yourself for wanting to watch something trashy that shows up. Watch it as many times as you like! But after you’re done watching it, remember to tap “not interested.” In my experience, the algorithm is quite responsive to this feedback, more so than simply scrolling past. But the hard part is remembering that this option even exists, especially because you weirdly have to tap “share” to access it. But hopefully, after the first few times you do it, it will become a habit.

Thanks to Roy Rinberg for feedback on a draft.

For more information on Arvind Narayanan’s Knight Institute work on algorithmic amplification, see his post introducing his project and our call for participation in “Optimizing for What? Algorithmic Amplification and Society,” a symposium to be held at Columbia University in April 2023.