Fundamentals

Surveillance Capitalism: What It Is and How It Works

By Daniel Sardá · Published on October 21, 2026

7 min read1,429 words

Surveillance capitalism describes an economic logic that turns behavioral data into predictions and opportunities for influence. Understanding it requires distinguishing observation, inference, personalization, and coercion.

An app recommends a song just when it seems there is nothing new left to hear. A store remembers a product viewed days earlier. A social network reorganizes content according to each user's pauses, clicks, and interactions. These experiences can be convenient, but they also reveal an economic shift: many companies do not merely provide a service; they learn from the behavior of the people who use it.

Surveillance capitalism is a critical framework for interpreting that shift. The scholar Shoshana Zuboff used the term to describe a logic of accumulation based on extracting data from human experience, processing it to predict behavior, and monetizing the results. Her academic formulation appeared in 2015, and the idea gained international attention with her 2019 book.

It is not, however, a neutral category or an uncontested description. It is a thesis about how part of the digital economy works and about the power that may arise from it. It is worth understanding the verifiable mechanism first, then assessing the reach of the critique.

Key idea: Surveillance capitalism does not simply mean that a company holds data. It describes an economic logic in which observing, inferring, and predicting behavior become a systematic source of value and power.

From a signal to a prediction

The process begins with signals that can come from different sources. Guidance from the European Data Protection Board distinguishes between data that are provided, observed, and inferred.

A person provides their age when creating an account. A platform observes that they clicked several bicycle ads. A system may then infer that they are interested in urban cycling. That last category is not a fact declared by the user; it is a probabilistic conclusion drawn by relating data.

Automated profiling organizes those signals to assess or predict such matters as interests, preferences, behavior, or location. It does not mean that a system knows a person intimately, nor that it is always right. It constructs categories and probabilities that can be used to rank content, select an offer, calculate risk, or decide which ad to display.

In Zuboff's terms, behavioral surplus includes data captured beyond what is necessary to provide or improve a service. Those data feed processes that generate estimates of future actions. The basic sequence can be summarized as behavioral signals, inferences, profiles, predictions, and commercial use.

Monetization does not require a company to sell a database to an advertiser. It may sell access to a segmented audience: the advertiser chooses certain characteristics, and the platform decides who sees the message. It may also use its predictions to improve recommendations, retain attention, or facilitate transactions within its own ecosystem.

What the concept includes—and what it does not

Calling every digital activity “surveillance capitalism” dilutes the concept. Several distinctions help preserve its usefulness.

Personalization alone is not surveillance. A service can tailor a function using limited information that is stored locally or knowingly supplied. The critique becomes more relevant when collection is extensive, opaque, and aimed at predicting or modifying behavior for economic purposes.

Behavioral advertising is not synonymous with the entire model. It is one of its most visible applications: ads selected from past activity or inferred profiles. But Zuboff advances a broader thesis about accumulation, knowledge, and power.

Influence is not coercion. Ranking options, highlighting a message, or taking advantage of a moment of vulnerability can steer a decision. That is not the same as removing a person's capacity to choose. Predictions alter probabilities; they do not deterministically control people.

Commercial surveillance and state surveillance are different. Companies commonly seek revenue, retention, or competitive advantage; the state has legal and coercive powers that a private actor lacks. The two may share technologies or exchange information, but conflating them prevents an assessment of their particular risks.

Useful distinction: An accurate recommendation does not prove manipulation. The relevant questions are what data made it possible, what was inferred, how transparent the process was, and whether the person had a real ability to accept or refuse it.

Everyday examples of the mechanism

A pixel embedded in a webpage can record that a browser visited a particular product. A software development kit integrated into an app can transmit usage events. Combined with activity within a platform, these signals can create segments such as “people likely interested in travel” or “users with purchase intent.”

These examples show why it is important to separate observed data from inferred data. Visiting a page happened; purchase intent is an interpretation. When a profile is used to decide what message someone receives, an uncertain inference begins to shape that person's informational environment.

Scale and context matter. Remembering a preferred language does not raise the same issue as tracking activity across services to infer sensitive traits. Nor is a voluntarily enabled option the same as a system that is difficult to detect or leave.

Privacy, autonomy, and power

The first concern is digital privacy, understood not only as secrecy but as the ability to decide who knows which aspects of one's life and for what purpose. A visible form allows a person to assess what they are giving up. Opaque profiling makes that assessment harder because a person may not know either the signals collected or the conclusions drawn from them.

This creates an information asymmetry: the organization knows much about the aggregate patterns of its users, while each user knows little about the models that classify them. That inequality of knowledge can become an inequality of power when it affects prices, opportunities, available information, or terms of access.

Personal autonomy is also at stake. Choosing freely does not require living without influences; every society contains persuasion, reputation, and advertising. The problem arises when influence is hard to recognize, rests on inferred vulnerabilities, or offers no practical way out.

From a liberal perspective, the answer is not to assume that users are unable to decide, nor to give the state unlimited access to the same instruments. It is to protect spaces of choice through general rules: transparency, meaningful consent, accountability for abuse, and clear limits on both private and public power.

Liberal criterion: Freedom of contract loses substance when one party cannot reasonably know what they are accepting. At the same time, protecting privacy must not become a justification for expanding state surveillance.

An influential thesis, not a settled verdict

Zuboff's theory has provided a powerful vocabulary for real practices of tracking, profiling, and segmentation. Yet some of its broader conclusions remain under debate. A systematic map published in 2024 found a field dominated by conceptual studies and an empirical base that remains limited. There is firm evidence that signals are collected and profiles are created; it is harder to show how much a particular intervention determines individual behavior or a social outcome.

There are also benefits that a complete critique should acknowledge. Advertising funds services that are free or inexpensive to access, and personalization can save time and surface relevant information. Recognizing those benefits does not resolve problems of opacity or consent. It does help avoid treating users as entirely passive subjects and every recommendation as a form of domination.

Competition can offer alternatives with less tracking or subscription-based models. But it does not always correct practices that consumers cannot observe or compare. Switching services is not easy either when accumulated contacts, incompatible formats, or the lack of equivalent options matter.

How to assess a particular practice

Rather than deciding whether the whole digital economy fits under a single label, it is better to ask precise questions: What data are necessary to provide the service, and which are collected in addition? What inferences are produced? Can the person understand and refuse the processing? Is use limited to the stated purpose? Are there real alternatives? Who is accountable if a profile causes harm?

These questions make it possible to distinguish useful personalization from disproportionate extraction, and a persuasive offer from a deceptive practice. They also guide institutions that protect privacy without banning innovation or concentrating still more information in the hands of the state.

The lasting contribution of the concept of surveillance capitalism is to make visible that data are not merely a technical resource. Their collection organizes relationships of knowledge and power. A free society needs markets capable of innovating, but also people able to understand, choose, and set limits. The challenge is not to eliminate all prediction; it is to prevent digital convenience from turning autonomy into a condition that no one can verify or negotiate.