How Algorithms Decide What We Watch
Examining The Influence Algorithm Recommendations Have On Our Content Consumption
How Algorithms Decide What We Watch
Examining The Influence Algorithm Recommendations Have On Our Content Consumption
We live in a world where algorithms curate much of our online experience. Content platforms (like Netflix, YouTube, Spotify, etc.) use algorithms to recommend content tailored to our tastes. When we see bad recommendations, we vaguely understand that the “algorithm” is to blame. However, if they do their job well, we barely notice them all. While they have transformed the content discovery process, making finding things we like easier, they also limit our options in subtle ways. Understanding how these algorithms work — and how they sometimes fail — can help us be more conscious of their influence. As the content we consume shapes who we are and what we believe, we need to consider what it means to have our content choices curated by code.
How Do Content Platforms Use Algorithms
They use algorithms to determine what content users see and how it is presented to them.
An algorithm is a set of instructions that a computer follows to solve problems, perform tasks, or make decisions. All content platforms rely on algorithms. These platforms collect massive amounts of user data, such as engagement metrics (likes, views, etc.), demographics, interaction history, etc. They use algorithms called recommendation systems to analyze user data to understand what users want. These algorithms are trained to understand their preferences, previous decisions, and characteristics. They evaluate how well content matches a user’s interests. They determine how content is filtered, ranked, selected, and recommended, making it easier for users to find and discover relevant content.
Algorithms operate on a self-reinforcing feedback loop. When people consume the algorithm’s recommendations, this gives the system “proof” that you genuinely like the content. It uses this as validation to provide more similar content.
— Gabrielle Wasco, How Spotify Standardizes Our Listening
Content platforms commonly use three types of recommendation systems:
Content-Based Filtering — These algorithms recommend content by analyzing the attributes of a piece of content (tags, descriptions, etc.) the user has previously liked or interacted with. When deciding what to show a user, the algorithm matches the attributes of past favorites with new content that shares the same characteristics.
Collaborative Filtering — These algorithms identify patterns in user behavior to group users with similar tastes. When deciding what content to show a user, they determine which group’s preferences align most closely with the users. Then they compare their behavior to the group’s to predict what content the user will like.
Knowledge-Based Filtering — These algorithms recommend content by matching item attributes directly with information provided by users (preferred genres, favorite creatives, etc.). When deciding what content to show a user, instead of depending on past behavior, they rely on the user’s stated preferences to suggest options that fit those specifications.
Why Content Platforms Rely on Algorithms
Algorithms help them cater to people’s tastes.
Content platforms need to understand and cater to the preferences of a broad spectrum of users. Algorithms can help them decipher people’s tastes. They help them objectively understand and predict what content people like, so they can provide it to them. Without algorithms, platforms would have to manually curate content for users, which is not efficient. Algorithms help deliver the most relevant content to each user. For example, Netflix uses algorithms to organize content into highly specific genres so it can recommend TV shows and movies curated to user’s tastes. This helps attract users, reduce churn, increase viewing time, and increase user satisfaction. This means these platforms waste fewer resources and generate more revenue.
Algorithms improve their user experience.
Content platforms have no shortage of content. However, it can be challenging for users to find the right piece of content. Navigating a massive selection can be frustrating and overwhelming. Evaluating the options and making a decision can present a significant cognitive burden. Algorithms can make this process a lot easier for users. They guide them directly to the content they will most likely enjoy, saving them time and effort. This reduces the friction involved, creating a better user experience. Instead of searching through the entire library, users just need to look at a much smaller selection of recommendations catered to their tastes. Features like Spotify's "Discover Weekly" or Netflix's "Because You Watched" are successful because users can immediately find something they like without endless searching.
Algorithms keep their users engaged.
Content platforms need to keep people interested and engaged. These platforms have massive content libraries, but all these options are meaningless if users don’t find anything they like. Most people don’t want to waste time searching for something interesting. Algorithms help platforms match users with the relevant content. Algorithms to keep users engaged by continuously curating content that is likely to appeal to users. Each user interaction (likes, views, comments, etc.) acts as a data point for the algorithm to further refine content recommendations. This ensures users consistently see content they like, so they keep coming back to the platform.
Algorithms help reduce user churn.
Content platforms are always competing for users. As user acquisition costs rise, retention becomes critical because reacquiring users can often cost five times more than the cost of keeping existing users. Users will not hesitate to switch if they feel like they are no longer getting any value out of a service. One of the top reasons users leave is a lack of use, which is most likely a result of them not finding interesting content. Personalized content recommendations can help them reduce this churn.
How To Improve Content Recommendations
Prioritize the right type of engagement.
Content platforms want to maximize engagement and time spent on platforms because this directly benefits their bottom line. However, users often get trapped in a filter bubble where they only see content that aligns with what they have already seen. This means users no longer see anything truly “new” because the algorithm will keep showing them content they are guaranteed to like. For example, if you watch several action movies in a row, the algorithm will keep recommending more action content, making it hard to explore other genres. This happens because of the feedback dynamics of recommendation systems.
Algorithms are designed to be efficient. However, they follow predetermined rules when deciding what to recommend. When content platforms fail to consider what meaningful user engagement looks like, algorithms will optimize recommendations for the wrong outcomes. This leaves users unsatisfied, defeating the purpose of personalization. For example, when people watch a movie, they might watch the whole thing, watch most of it, or watch only some of it. If the product team prioritizes the wrong variable (for example, views instead of viewing time) when training the algorithm, the algorithm will recommend the wrong content. Traditional engagement metrics often prioritize quantity over quality. Platforms should consider adopting value-based metrics that assess long-term user satisfaction.
Ensure users have a diverse selection of content.
Content platforms rely on algorithms to recommend content that matches users' tastes. In the short term, users can keep finding content they like. However, algorithms operate on a self-reinforcing feedback loop — each piece of recommended content the user consumes serves as confirmation to keep serving similar content. Therefore, as the algorithm continually refines recommendations the variety of options shown reduces, resulting in users seeing slight variations of content they have already consumed. For example, if a user keeps watching the suggested TV shows, over time the algorithm suggests shows that are more or less the same. This leads to people switching to a different platform — in hopes of finding “new” content.
These platforms don’t lack content. The problem is not that the algorithm restricts users, limiting opportunities to discover other interesting content. They are optimized for efficiency, which narrows content variety, resulting in user frustration and attrition. Sometimes keeping things interesting requires making algorithms less efficient by incorporating serendipity. This means adding some randomness to recommendations based on past user behavior. For example, if the user mostly listens to pop music, it might be useful to occasionally recommend songs from other genres. This creates some variety in recommendations. Another way to do this is by combining algorithmic and human-curated recommendations. For example, Spotify uses a hybrid recommendation system to balance “standard” recommendations with serendipitous content, creating richer user experiences.
Incorporate alternate ways to curate recommendations.
Streaming services have to figure out how to personalize recommendations without restricting their users. One part of this challenge is designing better algorithms. The other part is designing better platforms. There is no real way to “trick your algorithm” into giving you better recommendations. We really should not have to “fix” our algorithms. Instead, platforms should incorporate features that help users actively curate their recommendations.
Algorithm Feedback Mechanisms: Platforms could add the ability to directly “refresh” recommendations. Many platforms have feedback mechanisms for individual pieces of content (Netflix lets you like/dislike content, Amazon lets you hide a movie or series, etc.) but there’s no easy way to indicate that multiple recommendations are good or bad. It might be useful to have the option to generate new suggestions (like a refresh recommendations button) within the platform itself. This would provide another data point to indicate areas where the recommendation system may fall short and need adjustments.
Social Recommendations: Platforms could add the ability to share recommendations. Many platforms offer a way to create a watchlist but there’s no easy way to share this with people within the platform. Our social connections have always been a part of our natural content discovery process. People trust recommendations made by people they know. Incorporating this social element might open up new ways for users to find things they like. For example, HBO Max is exploring features (like friend-to-friend recommendations) that will connect users with other human recommendations.
Conclusion
Better recommendations help users find and enjoy more of a platform’s content.
Algorithms are now central to how content is consumed online. They help platforms personalize experiences at scale. However, many algorithms over-optimize content discovery, narrowing our exposure to diverse content. As the fight for our attention grows more intense, with platforms competing to win us over, people are more willing than ever to switch. Building better recommendation systems means taking a broader view of what value means to people. Our previous interests should not limit our current choices. Most platforms have enough interesting, compelling content. They just need to get better at helping people discover it. Instead of trying to help users find more options, platforms should think about how to help them find better options.
Thanks For Reading
References
Spotify | Algorithmic Effects on the Diversity of Consumption on Spotify
UniteSync | The Impact of Spotify Algorithms on Music Discovery
Carnegie Mellon AMT Lab | How Streaming Services Use Algorithms
The Conversation | How Netflix affects what we watch and who we are
The Influence of Social Media Algorithms on Consumer Buying Behaviour
Alexis Anzieu | Introducing Serendipity into Recommendation Algorithms
EY | How streaming services can drive profitability with customer centricity
Stanford University Human-Centered AI | When Algorithms Compete, Who Wins?
The Decisions Lab | Chained to the AlgoRhythm: How Spotify Standardizes Our Listening