How Recommendation Algorithms Work (and How to Take Back Control of Your Feed)
Why does TikTok, YouTube or Instagram seem to read your mind? A plain-English look at how recommendation algorithms decide what you see, and simple ways to reset your feed.
Information current as of November 2026 · How we research · Report an error
You watch one video about sourdough bread, and suddenly your feed is full of bakers. You pause on a sad post for a few seconds, and more sad posts appear. It can feel like the app is reading your mind. It isn’t. It is doing something simpler and more powerful: predicting what will keep you watching, based on what you and millions of people like you have done before.
This guide explains how recommendation algorithms work, what people mean by “feeding the algorithm”, and how to steer your feed back toward what you actually want.
What is a recommendation algorithm?
A recommendation algorithm (or recommender system) is software that picks, from a huge pool of content, the small number of items you are most likely to engage with, and puts them in order.
You meet them everywhere:
- Social feeds: TikTok’s For You page, Instagram Reels, Facebook, X.
- Video and music: YouTube’s home page and “Up next”, Netflix rows, Spotify’s Discover Weekly.
- Shopping: “Customers who bought this also bought…” on Amazon and AliExpress.
- News and search: Google Discover, news apps.
There might be millions of possible videos for you right now. The algorithm’s job is to choose the few dozen you’ll see first.
The basic recipe: three steps
The details are secret and change often, but most large platforms follow a similar pattern.
1. Gather candidates
The system first pulls a few thousand items that might interest you: posts from accounts you follow, videos similar to ones you liked, things that are popular in your country or language, and things people with similar tastes enjoyed.
2. Score and rank them
Next, a machine learning model predicts, for each candidate, how likely you are to do things like:
- watch it to the end, or for how many seconds,
- like, comment, share or save it,
- follow the creator,
- skip it quickly, hide it or report it.
Those predictions are combined into a score, and the candidates are sorted. If you are curious about the “learning” part, see What Is Machine Learning?.
3. Mix and filter
Finally, the platform adjusts the list: it removes content that breaks its rules, avoids showing five videos from the same creator in a row, and mixes in a few new or unexpected items to test whether your interests are changing. That exploration is why a random topic sometimes appears out of nowhere.
What signals does the algorithm watch?
Platforms use many signals. The most important ones are usually:
- Watch time and completion: finishing a video, or watching it twice, is a very strong signal.
- Quick pauses: even stopping your scroll for a couple of seconds counts.
- Interactions: likes, comments, shares, saves and follows.
- Negative signals: fast skips, “Not interested”, hiding a post or muting an account.
- Searches: what you type into the app’s search bar.
- Context: your language, rough location, device and the time of day.
- People like you: if many users who liked the same things as you also liked a video, you’ll probably see it too. This idea is called collaborative filtering.
A key point: you don’t have to like something for the algorithm to learn from it. Simply watching tells it a lot.
What does “feeding the algorithm” mean?
“Feeding the algorithm” means deliberately giving it the signals you want it to learn from. Creators use the phrase when they ask you to like, comment or watch to the end, because those actions push a post to more people.
As a viewer, you can feed it too, on purpose. Every video you finish, every search and every “Not interested” tap is a small vote for the kind of feed you’ll get tomorrow.
Why feeds can become a “rabbit hole”
Algorithms are very good at optimizing for engagement, and engagement isn’t always the same as what’s good for you. A few common side effects:
- Narrowing: the more you watch one topic, the more you get, until other interests disappear. Researchers and critics call extreme versions of this a “filter bubble”, although how strong the effect is in real life is still debated.
- Emotional content wins: posts that make people angry, scared or amazed often get more reactions, so they can spread further.
- Endless scrolling: autoplay and infinite feeds remove natural stopping points.
None of this means algorithms are evil. They also help you find useful tutorials, small creators and music you would never have discovered alone. The goal is to use them on purpose.
How to take back control of your feed
On any app
- Tap “Not interested” (or “Don’t recommend channel”) on things you don’t want. It’s more effective than just scrolling past.
- Search for what you do want and watch a few good videos fully. That quickly shifts your recommendations.
- Use the “Following” tab instead of the recommended feed when you only want people you chose.
- Unfollow or mute accounts that leave you feeling worse.
- Turn off autoplay where you can, and set a daily time limit in your phone’s Screen Time (iPhone) or Digital Wellbeing (Android) settings.
Curious how much time your feed really takes? Try our free Screen Time Calculator.
App-specific tools
- TikTok: Settings → Content preferences, where you can refresh your For You feed to start fresh and filter out keywords.
- YouTube: remove videos from your watch history or pause history altogether in your Google account. Removing a video you regret watching helps stop similar ones.
- Instagram: use Not interested on Explore and Reels, and check the “Suggested content” settings to snooze suggested posts.
- Facebook: use Show more / Show less on posts, and the Feeds tab to see only friends or favorites.
Menu names change often, so if you can’t find an option, search the app’s help center for “refresh feed” or “content preferences”.
Protect your privacy, too
Recommendations rely on data about you. Reviewing ad settings, limiting tracking and clearing cookies reduces how much is collected. See How Browser Cookies Work and How to Use AI Chatbots Safely.
The bottom line
Recommendation algorithms are prediction machines. They watch what you do, compare you with millions of other users and rank content by how likely you are to engage with it. You can’t switch them off completely, but you can train them: search for what you want, finish the good stuff, tap “Not interested” on the rest and use the tools each app gives you.
→ Related: AI for Beginners: A Complete Guide and Practical Ways to Use AI in Daily Life