recommendation system

in English: Recommender System

basicwidespread

A program that guesses what might interest you based on your past behavior and the decisions of people similar to you, and presents it to you.

In plain terms

Think of the salesperson at your favorite bookstore who has known you for years: they know what you bought last time and what others with similar tastes purchased. When you enter, they already reach for the shelf. The recommendation system does the same, but with millions of customers and millions of products at once.

It is used to select a few dozen films, songs, products, news items, or job ads from an overwhelming supply that you are most likely to click on. Without this, the offerings of a streaming service or online store would be unmanageable.

It differs from a search engine in that you do not need to ask: it recommends on its own. It is important to know that it does not necessarily look for what is best for you, but what the operator set as its goal. This could be a purchase, time spent on screen, or the number of clicks.

How it works

Data collection: the system records what you viewed, what you bought, what you clicked on, what you scrolled past, how long you stayed on a piece of content, and what you rated.
Finding similarities: it checks who behaves similarly to you and what they liked (this is called collaborative filtering). It also examines which content is similar to what you liked (this is content-based filtering).
Candidate selection: it quickly narrows down millions of possibilities to a few hundred promising ones.
Ranking: a machine learning model estimates for each candidate how likely you are to click or buy, and orders them accordingly. Business factors can also influence this, such as margin or paid promotion.
Feedback: what you do with offers becomes new data, so the system continuously learns from you.

An everyday example

You open Netflix in the evening, and the 'Recommended for You' row features crime series exactly like those you watched in recent weeks. The same happens when an online store displays: 'Customers who bought this also bought these.'

Who is affected

everyoneFrom social media news feeds to music playlists, recommendation systems largely decide what you see and what you do not.

marketerYour product or content's visibility depends on whom and how often the platform's recommendation system shows it. → Marketing

entrepreneur, executiveA well-configured recommendation engine in an online store can increase basket value and the number of returning customers.

parent, student, teacherIn children's video and social applications, the recommendation engine determines what content follows and how long it keeps them engaged. → Education & Family

A common misconception

MythThe recommendation system eavesdrops, which is why it knows what I talked about yesterday.

In realityFor targeted offers, analysis of your clicks, searches, purchases, and the behavior of users similar to you is usually sufficient. From these patterns, surprisingly much can be inferred about you.

Limits and risks

The recommender serves the operator's goal (such as watch time or sales), which does not always align with your interests.
It can easily trap you in a filter bubble: it shows more of the same, and you encounter less novelty or differing viewpoints.
It performs poorly with new users and new products because it lacks sufficient data (this is the cold-start problem).
Its operation requires a lot of personal data, and it can infer sensitive information from your behavior, such as your health status or financial situation.

Similar, but not the same

search engine – The search engine answers your typed question, while the recommendation system offers something without a request, based on your behavior.
targeted advertising – In advertising, an advertiser pays to appear before you, whereas a recommender selects from the platform's own offerings, although the two often blend in practice.
Personalization – Personalization is a broader term meaning any customization (for example, price, layout, email), while a recommendation system is one of its tools.

Related concepts

Broader

machine learning, Personalization, artificial intelligence

Narrower

collaborative filtering, Content-based filtering, Hybrid recommendation system

On the edge

filter bubble, search engine, targeted advertising, Profiling, embedding

A brief history

The term collaborative filtering emerged in 1992, associated with the Tapestry email filtering system developed by researchers at Xerox PARC. The topic gained major public attention in 2006 when Netflix launched a one-million-dollar competition to improve its recommendation algorithm.

Where things stand

By the end of 2026, this is a core element of almost every major online store, streaming service, and social platform. In the EU, the Digital Services Act (DSA) requires platforms to explain clearly the basis for their recommendations. The largest platforms must also offer a choice that does not rely on your personal profile. The new direction involves conversational recommendation solutions combined with large language models. Additionally, an AI agent increasingly selects offers on behalf of the user.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

YouTube to let viewers rewrite its recommendation algorithm with a text descriptionSep 24, 2026 · Marketing

Also known as

recommendation system · Recommender system · recommendation engine · recommendation algorithm · recommender system · recommendation system · recommendation engine · RecSys

Sources

Wikipedia: Recommender system

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.