A recommendation algorithm is the system an app uses to choose what to show you next: the videos in your feed, the shows on your home screen, the “you might also like” products. It makes its guesses from what you have watched, liked, searched for and skipped, and from what people with similar tastes enjoyed.
Picture a shop assistant who remembers everything you have ever bought, every shelf you lingered at and every item you put back. Each time you walk in, they lay out what they think you will pick up. That is helpful, and it also means you mostly see more of what you already chose.
In a sentence: “I watched one cooking video and now the recommendation algorithm fills my whole feed with recipes.”
What does recommendation algorithm mean?
The plain point: it is what people mean when they blame “the algorithm” for their feed.
An algorithm is a set of steps a computer follows to get a result. A recommendation algorithm’s result is a ranked list: out of everything it could show you, which items come first. Big apps build these systems with machine learning, which learns patterns from examples. The job is a prediction: what will you probably watch, click or buy next? YouTube says your likes help its system “try to predict the likelihood that you will be interested in similar videos in the future”.1
Where the word comes from
The older name was “collaborative filtering”. In a 1997 article called “Recommender Systems”, in Communications of the ACM, Paul Resnick and Hal Varian wrote that “The developers of the first recommender system, Tapestry, coined the phrase ‘collaborative filtering’”, and argued for a wider name: “We prefer the more general term ‘recommender system’”.2 The sources checked for this page do not say when “recommendation algorithm” itself came into use.
What does your feed learn from?
The plain point: mostly your own actions, plus a little about your device and settings.
Several big platforms publish the signals they use:
| Service | What it says it uses | Worth knowing |
|---|---|---|
| YouTube | Uses: watch history, searches, subscriptions, likes and dislikes, “Not interested” clicks and surveys. | Note: it says the surveys help it understand satisfaction, “not just watch time”.1 |
| TikTok | Uses: what you like, share, follow, comment on and create; captions, sounds and hashtags; language, country and device. | Note: it says it avoids two videos in a row with the same sound or creator.3 |
| Netflix | Uses: your viewing and ratings, members with similar tastes, details of each title, time of day and device. | Note: it says it does not use age or gender.4 |
“Members with similar tastes” is a classic trick: if people who loved the same three shows as you also loved a fourth, you will probably be offered the fourth.
How it shapes what you see
The plain point: it learns from you, and then you learn from it.
Every click is a vote for more of the same. That is why a feed can drift towards one topic after a few evenings of watching it. Platforms say they push back on this: TikTok, for example, says it sometimes shows videos outside your usual interests to help you “discover new creators, and experience new perspectives”.5
Your feed is a mirror of your recent clicks, not a picture of the world. If it feels narrow, feed it different signals.
How to steer your feed
- Use the buttons: “Not interested” and “Don’t recommend channel” are signals the systems say they read.
- Clear the trail: YouTube says you can turn off and delete your watch history.1
- Ask for a plain feed: in the EU, large platforms must offer a feed that is not personalised.6
- Go looking: searching for and watching something new gives the system fresh signals.
Often confused with
- Search: in search you ask; in a feed the app decides for you. Some search tools find matches by meaning; see semantic search meaning.
- A chronological feed: newest first, no guessing involved.
- AI chatbots: a chatbot writes a new answer; a recommendation system picks from things that already exist.
Questions people ask
How does the TikTok algorithm work?
TikTok says its For You feed is driven by what you interact with, details of the videos themselves, and your device and account settings, with strong signals, such as watching a longer video to the end, counting for more than weak ones.3 It also says it breaks up repeats and mixes in some new topics.
Can I turn off the algorithm?
Partly. In the EU, large platforms must let you choose a feed that is not personalised.6 Elsewhere it depends on the app; YouTube, for example, says you can turn off and delete your watch history.1
Learn more on this site
- The basic word: algorithm meaning.
- Meaning as numbers: embedding meaning.
- How AI nudges you: how AI influences you back.
- The words: the AI glossary.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What does a recommendation algorithm mostly learn from?
- Your own actions, like what you watch, like and skip
- A fixed list of what is popular around the world
- Your age and gender, which every app must use
Show the answer
Answer: Your own actions, like what you watch, like and skip. The page says feeds learn mostly from your own actions, plus a little about your device and settings.
2. Why can a feed drift towards one topic after a few evenings?
- The app runs out of other videos to show you
- Every click works like a vote for more of the same
- The feed is always sorted newest first
Show the answer
Answer: Every click works like a vote for more of the same. The page says every click is a vote for more of the same, so watching one topic for a while can narrow the feed.
3. How is a recommendation system different from a chatbot?
- It writes a brand-new answer for each person
- It only works when you type in a question
- It picks from things that already exist
Show the answer
Answer: It picks from things that already exist. The page says a chatbot writes a new answer, while a recommendation system picks from things that already exist.
Related terms: Prompt chaining · Prompt engineering · Reasoning model · Role prompting · Sentiment analysis · Sycophancy
- 1YouTube, How YouTube recommendations work, undated help page, read at source 5 Oct 2026“The videos you’ve liked helps our system try to predict the likelihood that you will be interested in similar videos in the future.” ↑a“What you search on YouTube influences future recommendations.” … “Videos you mark as ‘Not interested’ helps to inform what to avoid recommending in the future.” … “User surveys that ask you to rate videos that you watched helps the system understand satisfaction, not just watch time.” ↑byou can “turn off and delete your watch history” if you prefer not to have recommendations on the homepage. ↑cyou can “turn off and delete your watch history” if you prefer not to have recommendations on the homepage. ↑d
- 2Social Science Research Council, Just Tech, citation record for Resnick and Varian, “Recommender Systems”, which quotes the article (Communications of the ACM, 1997), read at source 13 Oct 2026“The developers of the first recommender system, Tapestry [1], coined the phrase “collaborative filtering”” and “We prefer the more general term “recommender system” for two reasons.” The original article was not read: the publisher’s page refused the request on 13 Oct 2026. ↑
- 3TikTok, How TikTok recommends videos #ForYou, 18 Jun 2020, read at source 5 Oct 2026user interactions such as “the videos you like or share, accounts you follow, comments you post, and content you create” … “your For You feed generally won’t show two videos in a row made with the same sound or by the same creator” ↑a“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”; device and account settings “receive lower weight in the recommendation system relative to other data points.” ↑b
- 4Netflix, How Netflix’s Recommendations System Works, help page, read at source 5 Oct 2026“other members with similar tastes and preferences on our service” … “The recommendations system does not include demographic information (such as age or gender) as part of the decision making process.” ↑
- 5TikTok, How TikTok recommends videos #ForYou, 18 Jun 2020, read at source 5 Oct 2026, which says the system sometimes shows videos outside a user’s expressed interests to help them “discover new creators, and experience new perspectives”.↑
- 6European Commission, The Digital Services Act, updated 1 Oct 2026, read at source 5 Oct 2026“On large platforms with over 45 million monthly users in the EU, you can now opt for non-personalised feeds.” ↑a“On large platforms with over 45 million monthly users in the EU, you can now opt for non-personalised feeds.” ↑b
The comparison on this page (a shop assistant with a perfect memory), the “similar tastes” example, the point that feeds drift and the contrasts with search and chatbots are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK