The Hidden Cost of "Free"

How ad-supported business models extracted more than money from users

There is a phrase that spread quietly through the technology industry sometime in the mid-2000s, eventually achieving the status of an axiom: if you're not paying for the product, you are the product. It was pithy. It was widely shared. And for a long time, it was accepted as a sufficient explanation of how the ad-supported internet worked — a knowing shrug that acknowledged the trade-off without examining its full dimensions.

The axiom, it turns out, was incomplete. The exchange was considerably more lopsided than it implied, and the currency being extracted from users was considerably more valuable than it first appeared. "Free" services cost their users their attention, their data, their time, their psychological wellbeing, and in some cases their grip on reality. The accounting, done honestly, does not flatter Silicon Valley.

Understanding how this happened requires a short trip back to the early 2000s, when a search engine was trying to figure out how to pay its bills.

The Accidental Invention

Google's original sin — if one can call it that — was more accident than design. The company had amassed vast stores of behavioral data as a byproduct of operating its search engine. Users queried, clicked, lingered, backtracked. All of that activity left a trail, and somewhere around 2001, according to Harvard Business School Professor Emerita Shoshana Zuboff, Google's engineers discovered that surplus behavioral data could be used not just to improve the product, but to predict what users would do next — and sell those predictions to advertisers.

This was new. Not advertising itself, which is as old as commerce, but the specific logic of packaging human behavior into forward-looking predictions and selling them in what Zuboff calls "behavioral futures markets." In her landmark 2019 book The Age of Surveillance Capitalism, she describes the resulting economic system as one that "unilaterally claims human experience as free raw material for translation into behavioral data." The data is processed, the predictions are manufactured, and those predictions are sold to anyone willing to pay — brands, political campaigns, whoever has the budget.

Facebook, which arrived a few years after Google had already laid the template, adopted and accelerated the model. By the early 2010s, both companies were not merely observing what users wanted; they were engineering environments designed to maximize how long users stayed and how much behavioral data they generated. The more engaged the user, the more data. The more data, the more precise the prediction. The more precise the prediction, the higher the price an advertiser would pay.

This logic had a consequence that is now well understood but was poorly appreciated in the moment: the platforms became structurally incentivized to make their products as consuming as possible, regardless of whether that consumption was good for the people doing the consuming.

Attention as a Finite Resource

The attention economy — the idea that human attention is a scarce commodity that can be captured, monetized, and competed over — was not invented by the ad-supported internet. Newspapers, radio, and television had all operated on variations of the same model: gather an audience, sell access to it. But the digital version of this arrangement was qualitatively different in several important ways.

First, the feedback loop was instantaneous and extraordinarily fine-grained. A television network could tell you how many people watched a program; a social media platform could tell you exactly which posts held a user's eyes for four seconds longer than average, which topics triggered emotional responses, and which interface designs increased the likelihood of a compulsive return visit. This precision enabled optimization at a scale and speed no previous medium could match.

Second, the platforms were not passive carriers of content — they were active curators. Algorithms determined what users saw, in what order, and with what frequency. The governing criterion for those decisions was engagement, which in practice meant content that provoked strong reactions: outrage, anxiety, desire, envy. A 2019 study cited by the Attention Economy entry in Wikipedia found that when ad-supported platforms optimized for engagement alongside advertising, the financial arrangement actively incentivized the spread of disinformation — because false or emotionally provocative content reliably outperformed accurate, measured reporting in generating the reactions that kept users on-platform.

Third, the competition for attention was not bounded by physical space or broadcast schedules. It was available everywhere, at all hours, on a device that most adults and a growing number of children carried in their pockets. The result was an unprecedented colonization of idle time — commutes, meals, quiet moments before sleep — that had previously been available for reflection, conversation, or simply rest.

The aggregate effect was enormous. By 2024, according to data from the American Psychological Association, teenagers in the United States were spending an average of nearly five hours per day on social media alone. That is roughly one-third of their waking hours directed toward platforms whose economic interests were structurally misaligned with their wellbeing.

The Psychological Toll

The mental health consequences of this arrangement are now documented with enough consistency across enough independent studies to constitute something close to scientific consensus, at least in broad strokes.

A research study of American teenagers ages 12 to 15 found that those who used social media for more than three hours daily faced twice the risk of negative mental health outcomes, including depression and anxiety symptoms, compared to lighter users — findings reported by Yale Medicine. A 2022 meta-analysis examining the dose-response relationship between social media use and depression in adolescents found a statistically significant association that increased with time spent on platforms. A 2024 Pew Research Center survey found that 45 percent of American teens — nearly half — described their own social media use as excessive, up from 27 percent the year before.

What makes these statistics particularly damning is not the correlations themselves, which some researchers still debate in terms of causality, but what the platforms knew and chose not to act on.

In September 2021, The Wall Street Journal began publishing a series of articles based on internal Facebook documents leaked by former product manager Frances Haugen. The documents — which became known as the Facebook Papers — revealed that Meta's own internal research had found that 32 percent of teen girls said that when they already felt bad about their bodies, Instagram made them feel worse. A separate internal study found that 17 percent of teen girls reported that Instagram contributed to their eating disorders, and that 13.5 percent of U.K. teen girls said the platform worsened suicidal thoughts.

Facebook knew. The company had commissioned the research, read the findings, and — according to Haugen's congressional testimony — repeatedly chose not to act on them in any meaningful way because doing so would have required redesigning the engagement-maximizing features that generated its revenue. As Senator Richard Blumenthal noted during Haugen's Senate testimony, Facebook had specifically targeted teens and preteens through Instagram in order to drive up numbers for advertisers and boost profits for shareholders.

This was not a bug or an oversight. It was a direct consequence of the model's incentive structure. Platforms designed to maximize engagement are, by definition, designed to exploit psychological vulnerabilities — the same mechanisms that make social comparison, fear of missing out, and compulsive validation-seeking so effective at keeping people scrolling.

The Data That Was Never Yours

Beyond attention, the ad-supported model extracted something else: personal data, at a scale and intimacy that users had no meaningful framework to evaluate when they clicked "agree" on terms of service agreements that ran to tens of thousands of words.

What kinds of data? Demographics and behavioral patterns, obviously. But also location history, browsing habits off-platform through tracking pixels and cookies, contact lists, relationship graphs, political affiliations implied by the content engaged with, emotional states inferred from typing patterns, and increasingly, voice and image data as platforms expanded into those modalities. As Zuboff describes it, the most predictive behavioral data ultimately comes from the most intimate dimensions of human experience — not just what people do, but who they are.

This data was monetized in two principal ways. First, it was used internally to serve more precisely targeted advertising. Second, it was sold or shared with third parties — often, as a 2017 Columbia Law Review analysis noted, without users' informed consent in any meaningful sense of that phrase, since users "cannot reasonably estimate the marginal disutility that particular instances of data collection impose on them."

The scale of the resulting industry is staggering. Google's advertising revenue reached $237.8 billion in 2023. Meta's reached $131.9 billion the same year. Together, the two companies accounted for roughly 57 percent of all digital advertising revenue globally. The prize at the center of this empire was the behavioral data of billions of people who believed they were simply using a free service to search for information or stay connected with friends.

The wealth concentration this produced is worth sitting with. Google and Meta became two of the most valuable companies in human history largely by harvesting surplus behavioral data from users who received no financial compensation for the raw material they provided and no meaningful disclosure of how it would be used. The asymmetry is as complete as asymmetries get.

The Architecture of Engagement

There is one more cost that operates at a level more structural than individual psychology or data privacy — and it may ultimately prove the most consequential.

When platforms optimize for engagement above all else, and when the content that best drives engagement tends to be emotionally provocative rather than intellectually rigorous, the information environment that emerges is systematically distorted. Outrage outperforms nuance. Conflict outperforms consensus. Extremity outperforms proportion.

This is not a theoretical concern. Research cited by the Mercatus Center has found evidence that a considerable share of programmatic advertising revenue flows to sites presenting misinformation and fake news, precisely because those sites are effective at generating the user engagement that justifies the advertising spend. The attention economy, in this sense, became a subsidy mechanism for disinformation — not because the platforms set out to fund it, but because their economic logic made it the path of least resistance.

Before algorithmic feeds and behavioral targeting became dominant, advertising strategies often relied on long-term brand awareness built through association with quality editorial content. The ad-supported internet effectively dissolved that alignment. A brand's ad could appear next to content that inflamed, misled, or radicalized with equal efficiency, because the targeting was on the user, not the context. What the content did to the user's understanding of the world was not part of the economic calculation.

The cumulative effect of years of this environment on public discourse — on the ability of democratic societies to deliberate, find shared facts, and resolve disagreements through reason rather than escalating outrage — is difficult to quantify but hard to dismiss. The attention economy did not create political polarization or epistemic fragmentation, but it built the most efficient delivery mechanism for both that has ever existed, and then it ran it at scale for two decades.

What "Free" Actually Cost

None of this means the ad-supported internet produced no value. Search engines, social networks, maps, email, video platforms — these tools have genuinely changed how people navigate the world, maintain relationships, and access information. The exchange, in its early form, seemed fair enough: useful services in return for exposure to advertising. The problem was that "exposure to advertising" proved to be a description that radically understated what was actually being taken.

The real accounting looks something like this. Users surrendered their behavioral data permanently, without meaningful compensation or recourse, to build prediction engines of extraordinary precision. They surrendered hours of daily attention to platforms designed by behavioral scientists to be as difficult to disengage from as possible. Adolescents, in particular, surrendered developmental years to environments that the platforms' own researchers had identified as damaging to body image, mental health, and self-worth — and that the platforms chose not to redesign because redesigning them would have cost revenue.

And society at large surrendered something harder to quantify: a shared information environment with at least some connection to a common set of facts, replaced by algorithmically personalized realities optimized for emotional engagement rather than accuracy.

The phrase "if you're not paying for the product, you are the product" was always too narrow. It implied a simple barter — your eyeballs for a service. What actually happened was more extractive, more intimate, and considerably more consequential. Users were not the product. They were, as one pointed critic of Zuboff's framing observed, the site of raw material extraction — and the extraction went considerably deeper than anyone who signed up for a free email account in 2004 had any reason to anticipate.

The bill, as it turns out, is still being tallied.