How algorithms decide what appears in your feed

Every time you open Instagram, TikTok or YouTube, a decision has already been made before your first scroll: what will appear, in what order and how often.

This decision does not come from a human editor or a chronological order of publications, but comes from a set of automated systems that have learned, over time, what captures your attention. The learning obtained is used to create, in fractions of a second, a personalized feed for each user.

For Daniel Delgado, a specialist in artificial intelligence, understanding how these systems work is more than a technical curiosity. It's a way to regain, at least in part, control over what you consume. "Those who understand the game gain real control. Those who don't understand only gain the illusion of choice", he states.

What actually is the algorithm?
The first thing the expert clarifies is that the algorithm is not a single program. "It's a set of rules and models that learn from data and that, together, decide the order and priority of what each person sees every time they open or update the application", he explains.

The process takes place in three stages. First, the system brings together everything that could be shown to that user at that moment. It then filters this volume of content until a group of candidates considered relevant remains. Finally, it gives a score to each candidate and organizes them from most to least likely to capture that specific person's attention.

But Daniel is keen to point out a tension that often goes unnoticed: "The platforms' stated objective is to deliver relevance and a good experience. The practical objective, which sustains the business, is to capture and maintain your attention for as long as possible, because it is the time of use that generates revenue from advertisements." Relevance and retention, as he explains, don't always point in the same direction.

The four factors that shape the feed
When it comes to personalization, Daniel organizes the factors into four blocks. The first is the user's own recent activity: what they liked, commented, saved, rewatched, who they followed and, most importantly, how long they spent on each content.

The second are the characteristics of the content itself: theme, audio, caption, format and the history of performance of that post with people with a similar profile.

The third is the relationship between the user and the poster: exchange of messages, frequent comments, profile visits.

The fourth is context: language, location, device type and time.

Among all these signs, the expert highlights a principle that he considers central: "The algorithm gives more weight to what you do than to what you say you like. You can say that you love finance content, but, in practice, you watch more humor videos, it is humor that will dominate your feed."

Liking, commenting, sharing: which is worth more?
One of the most common questions about how the algorithm works is whether all forms of interaction have the same weight. The answer, according to Daniel, is no. And the recent trend points to a clear hierarchy, with attention span at the top.

"Everything you do is a clue that tells the algorithm 'this caught my attention'. Liking is one clue. Saving content is another. Watching the same video two or three times is another, and one of the strongest", explains the expert.

Other interactions are worth depending on the effort they require from those who act. Likes are cheap, cost little and are worth little. The comment is intermediate, and those that generate conversation weigh more than an isolated emoji. Sharing is the most expensive: when someone sends content to someone else, they put their own reputation at stake by recommending it, which makes it one of the strongest signals for reaching new audiences.

Daniel still draws attention to the clues that most people don't even realize he's giving. "The platforms record whether you rewatched, whether you paused in the middle, whether you swiped at the time and even the speed at which you scrolled the screen," he says. Rewatching, for example, is one of the most telling signs of genuine interest.

Is the cell phone "listening" to conversations?
One of the most widespread beliefs about social networks is that applications listen to users' conversations to display related advertisements. The expert's answer, in line with what digital security companies like Kaspersky and AVG and vehicles like CBS News and Washington Post concluded after technical tests, is straightforward: no, at least not in the way most imagine.

But the explanation for the feeling of being heard is, according to the expert, more disturbing than the microphone theory itself. The first reason is the accuracy of data tracking: companies build a detailed profile of the user based on browsing history, location, behavior in applications and crossing between devices, without the need for audio.

The second is proximity: as the applications know the location, the system notices when two people on the same network are close and uses the interest of one as an indication of the interest of the other.

The third is psychological. "There is a phenomenon called illusory correlation, which is our tendency to notice things that coincide with something we have just thought about or talked about. You forget the thousand times you talked about a subject and nothing appeared, but you vividly remember the time the announcement came up shortly after," explains Daniel.

How to teach the algorithm to work in your favor
By knowing the rules of the game, you can influence what the algorithm learns about you. "The algorithm learns from what you do, so act with intention," explains Daniel.

In practice, this means watching until the end and saving the content that really interests you, but also actively acting on what you don't want to see: swiping, hiding or marking "I'm not interested" teaches the system from the negative side, which is just as effective as liking.

Commenting and sharing on desired topics are the strongest signals to increase this type of content in the feed. The platforms have also expanded their control tools: Instagram, for example, now allows the "reset" of suggested content in the settings, erasing the history used by the algorithm and starting over from scratch in one or two days.

"A practical tip: right after resetting, it's worth spending 15 to 20 minutes engaging with just the right content, to train the new algorithm in the direction you want", advises the expert.

What changes with the advancement of AI?
For the future, the dominant direction identified by the expert is hyper-personalization: feeds adjusted to each person's behavior, context and intention, with systems processing millions of data in real time.

Artificial intelligence is also already involved in creating and adapting content itself: Meta, through the Advantage+ package, already allows advertisers to automatically generate images, videos and texts adjusted for different audience profiles. By 2025, more than four million advertisers were already using these tools.

About user control in this scenario, the expert states: there are more explicit tools available, but more advanced systems also tend to be more difficult to understand and more persuasive.

"The most likely scenario is a paradox: technically, more control options, but requiring more knowledge and more active intention to actually use them", he concludes.

Is your Android spying on you? 5 settings to protect yourself

Source: CNN