Do Invisible Algorithms Shape What You See Online?
Short answer: yes. The recommendation, ranking and filtering systems built into the platforms we use every day decide, for each of us individually, which content appears and in what order. They operate quietly in the background, and in doing so they influence not only what we watch and read, but also, over time, our opinions and choices.
Most people never encounter these systems directly. There is no visible menu of decisions, no notice explaining why one video was surfaced and another buried. Yet the effect is continuous. This article explains, in plain terms and with reference to recent research, how these invisible algorithms work and why they matter.
What are the “invisible” algorithms?
Three broad types of system do most of the work. Recommendation algorithms suggest what to watch or read next — the mechanism behind YouTube’s autoplay, TikTok’s feed and Netflix’s home screen. Ranking algorithms order search results and social feeds by predicted relevance and engagement, determining what is seen first and most often. Filtering algorithms decide what to leave out, sparing us from overload but also narrowing the range of sources we encounter.
Together, these systems personalise the digital environment for each user. They are also, in most cases, proprietary and opaque: the companies that build them rarely disclose how decisions are made, which makes independent scrutiny difficult.
How do they build a personal “bubble”?
These systems learn from behaviour. They combine explicit signals — a search, a like, a subscription — with implicit signals that users rarely realise they are giving off: how long they pause on a post, what they scroll past, what they replay. From these traces, the algorithm assembles a personalised selection, a kind of bubble fitted to each individual.
The concern is that this personalisation tends to reinforce what a person already seeks out or already believes, reducing the diversity of viewpoints they are exposed to. Research quantifying online information exposure has shown that ranking and personalisation can distort our sense of what is popular and narrow the range of sources we see (Nikolov et al., 2019, JASIST). The magnitude of this effect is still debated among researchers, but the mechanism itself is well documented.
Why does optimising for engagement matter?
There is a second, more robust finding. Because many of these systems are designed to prolong the time users spend on a platform, they tend to favour content that provokes a reaction — material that is emotionally intense or divisive — because that is what keeps people engaged.
A 2025 study in PNAS Nexus found that ranking content by engagement can amplify divisive political material beyond what users themselves say they want to see (Milli et al., 2025). Work in the Journal of Public Economics has likewise linked engagement-based ranking to the spread of misinformation and polarisation (Germano et al., 2026). The practical consequence is that neutral or evidence-based content tends to gain comparatively less visibility.
Is this only about social media?
No. The same logic applies wherever content is distributed at scale. Streaming services such as Netflix use recommendation systems to shape what we watch; search engines rank what we find; shopping platforms order what we are offered. The mechanics differ, but the underlying principle — personalised, automated selection optimised toward a commercial objective — is shared across the ecosystem.
Why does algorithmic literacy matter, especially for young people?
If we cannot see how these systems work, we are poorly placed to judge what they show us. This is where algorithmic literacy — understanding that a feed is a curated, automated selection rather than a neutral picture of reality — becomes important.
The research points to two things worth noting. First, this literacy improves informed internet use: people who understand how algorithms shape their feeds are better able to interpret and question what they see (Gruber & Hargittai, 2023, Big Data & Society). Second, it is unevenly distributed. Algorithmic knowledge tends to be higher among younger, more educated and more intensive users, while less-privileged groups are less able to detect and respond to bias — a gap that can deepen existing inequalities (Cotter & Reisdorf, 2020; Chung & Wihbey, 2024).
Because young people are growing up inside these systems, building their literacy early is a reasonable priority — with an important caveat the evidence makes clear: understanding alone does not guarantee action. Many users remain passive even after they grasp how the systems work.
The takeaway
The content on our screens is the product of an automated choice — one made for us, not by us. Recognising this is the first step toward using these platforms more deliberately: seeking out a wider range of sources, treating “popular” as a signal rather than a verdict, and remembering that what we do not see has been filtered out just as deliberately as what we do.
Frequently asked questions
What is an algorithmic filter bubble?
An algorithmic filter bubble is the narrowed information environment that results when recommendation and ranking systems repeatedly show a user content aligned with their past behaviour and apparent preferences. The effect can reduce exposure to differing viewpoints, although researchers continue to debate how large it is in practice.
Do recommendation algorithms cause political polarisation?
Recommendation algorithms do not cause polarisation on their own, but ranking content by engagement can amplify divisive material because such content provokes stronger reactions. Studies published in PNAS Nexus (2025) and the Journal of Public Economics (2026) link engagement-based ranking to increased polarisation and the spread of misinformation.
Which platforms use these algorithms?
Recommendation, ranking and filtering algorithms are used across most large digital platforms, including YouTube, Instagram, TikTok, search engines, online shops and streaming services such as Netflix. Any service that personalises what content appears, and in what order, relies on some form of automated selection.
Can I find out why a platform showed me something?
Some platforms offer transparency features, such as “Why am I seeing this?” labels. Research suggests these tools can improve understanding, but their usefulness depends on the user’s existing algorithmic literacy, and the underlying systems remain largely proprietary and difficult to scrutinise independently.
What is algorithmic literacy?
Algorithmic literacy is the understanding that an online feed is a curated, automated selection rather than a neutral picture of reality, together with the ability to interpret and question what that selection presents. Research indicates it supports more informed internet use, but that it remains unevenly distributed across social groups.
How can I reduce the effect of algorithmic personalisation?
Practical steps include consulting a deliberately wide range of sources rather than relying on a single feed, treating “popular” or “trending” as a signal rather than a verdict, using available transparency and feed-control settings, and periodically reviewing recommendations critically. No single measure removes personalisation, but combined they broaden exposure.
About the author. Miguel Cachulo Pereira is an Assistant Professor at ISCA – University of Aveiro and a researcher at the CEOS.PP and GOVCOPP research centres, where his work focuses on MarTech and digital marketing. His publications include research on algorithmic bias in marketing. See his full profile on ORCID, Google Scholar and Scopus.