Last Tuesday, sitting in a café in Tiong Bahru, I tapped “I agree” on a cookie consent banner without reading it. Then I did it again on a different site. Then a third time. By my rough estimate, I’ve done this somewhere around 50,000 times in my adult life. I’ve probably spent more cumulative seconds agreeing to terms I haven’t read than I’ve spent learning to cook. And I say this as someone who writes about power for a living.
That evening, I decided to do something I’d never actually done: trace, as precisely as I could, the chain of value that activates every time someone taps that button. Who gets paid? How much? Through what mechanisms? I spent weeks pulling at this thread, talking to adtech engineers, reading broker filings, cross-referencing data supply chain maps published by researchers at Cracked Labs and the Irish Council for Civil Liberties. What I found wasn’t a conspiracy. It was something more durable than that: a class system, encoded in infrastructure, operating at a speed and scale that makes it functionally invisible to the people who feed it.
The tap itself: what happens in 300 milliseconds
When you tap “I agree” on a typical European or American news site, a real-time bidding auction begins. Your device sends a bid request containing what the industry calls “signals”: your approximate location, device type, browsing history (if third-party cookies are still active), and increasingly, probabilistic identity graphs that link your behavior across apps and websites even without cookies. This bid request goes to an ad exchange, which broadcasts it to dozens, sometimes hundreds, of demand-side platforms in under 100 milliseconds.
The winning bidder gets to show you an ad. The publisher gets a fraction of the ad spend. And somewhere between 30 and 70 percent of the money that the advertiser originally paid is absorbed by intermediaries: the exchange, the DSP, the data management platforms, the verification vendors, the identity resolution firms. A 2020 study by PwC and ISBA found that for every pound spent by an advertiser in programmatic display, only 51 pence reached the publisher. Fifteen percent of the spend was completely untraceable, lost in what the researchers called an “unknown delta.”
I want to sit with that phrase for a moment. Unknown delta. Fifteen percent of billions of dollars flowing through a system so opaque that forensic accountants commissioned specifically to trace the money couldn’t tell you where it went. This is the plumbing of the modern internet. This is what your tap funds.
The consent economy as class architecture
Here’s what struck me as I mapped this system: the distribution of who benefits and who provides raw material mirrors, almost perfectly, older structures of extraction. At the bottom, you have billions of people generating behavioral data through the simple act of living digitally. They receive a “free” service (email, social media, search) in exchange for data that they cannot price, cannot audit, and cannot meaningfully withdraw. In the middle, you have publishers and app developers who serve as the collection points, scraping enough ad revenue to survive but rarely to thrive. At the top, you have a small number of infrastructure firms: the exchanges, the clouds, the identity graph operators, the platforms that own both the demand and supply side of the marketplace.
The language of consent obscures this. “I agree” sounds like a negotiation between equals. In practice, it’s a legal fiction that converts asymmetric power into the appearance of choice. The philosopher Carissa Véliz has written persuasively about this: consent in the data economy functions the way “voluntary” labor agreements functioned in company towns. The alternative to agreement is exclusion from participation in modern life.
I think about this dynamic often when I look at how technology reshapes the terms of belonging. The question of who gets to participate, and on what terms, is ultimately a question about class, even when it’s dressed in the language of innovation and user experience.
The brokers you’ve never heard of
The most profitable layer of this system is the one furthest from public view. Companies like Acxiom (now LiveRamp), Oracle Data Cloud (which quietly shut down its advertising division in 2022 after regulatory pressure), Lotame, and dozens of smaller data brokers operate in a space that most consumers don’t know exists. They aggregate behavioral data from consent flows, merge it with offline purchasing records, voter files, and property data, and sell packaged audience segments to advertisers.
The segments have names that sound clinical until you think about what they represent. “Financially stressed.” “Expecting parent.” “Chronic condition: diabetes.” “Likely subprime borrower.” Each of these is a dossier on vulnerability, sold to the highest bidder at scale.
I obtained pricing sheets from two mid-tier data brokers (neither of which I’ll name, because the sheets were shared with me on background). The cost of reaching 1,000 people identified as “high-net-worth individuals” was roughly four times the cost of reaching 1,000 people identified as “budget-conscious households.” This makes market sense. But consider what it means structurally: wealthy people’s attention is literally valued more highly in the system’s own accounting. Poor people’s data is cheaper to buy because the products marketed to them have lower margins. The system doesn’t just reflect class: it prices it, reinforces it, and automates it at a speed no human institution could match.
Where the money actually accumulates
Follow the money far enough and you reach a familiar set of names. Google and Meta together capture over 50 percent of global digital ad revenue. Amazon is the fastest-growing third player. These companies operate as both marketplace and participant. Google runs the auction (through Ad Manager and AdX), represents the buyers (through DV360), represents the sellers (through AdSense and Ad Manager), and owns the verification and measurement tools. The conflict of interest is structural, and it has been the subject of antitrust litigation in both the US and EU.
But here’s what most coverage of adtech monopoly misses: the real power isn’t in the ad revenue itself. The real power is in the data layer that the ad system generates. Every auction, every bid request, every conversion pixel is a signal that feeds machine learning models. These models power product recommendations, content ranking algorithms, predictive analytics tools, and increasingly, enterprise AI services sold to governments and corporations. The advertising system is, in a sense, a subsidy mechanism. Consumers provide free training data through their behavior. The ad economy monetizes their attention. And the intelligence extracted from both funds the next generation of AI infrastructure.
I’ve been thinking about this feedback loop in the context of broader conversations about how artificial intelligence is reshaping the economy. The people tapping “I agree” are, in a very literal sense, performing unpaid labor that trains the systems increasingly likely to automate their jobs. The class dynamics here are not subtle. They are just fast.
The European experiment and its limits
Europe’s General Data Protection Regulation was supposed to change this. In many ways, it has. GDPR created the consent banners that now wallpaper the web. It gave individuals the right to access, delete, and port their data. It imposed fines that have, in a few notable cases, reached into the hundreds of millions of euros.
But GDPR also, paradoxically, entrenched the system it was designed to reform. The consent banner itself became a dark pattern arms race. Researchers at Ruhr University Bochum found that fewer than 12 percent of consent banners on major European websites met the regulation’s actual requirements. “Reject all” buttons were hidden, minimized, or required multiple clicks while “Accept all” was bright, prominent, and one tap away. The regulation mandated consent. The industry optimized for manufacturing it.
More fundamentally, GDPR’s individual-rights framework treats the problem as a series of bilateral negotiations between a user and a service. The actual power asymmetry is systemic, not individual. My ability to refuse consent on a single website does nothing to address the identity graphs, the broker networks, the auction infrastructure. It’s like giving a factory worker the right to individually negotiate their wages while leaving the entire structure of capital ownership untouched. The right is real. Its power to change outcomes is limited.
What I’m actually arguing for
I want to be honest about my own position here. I run a media company. Brown Brothers Media operates within the same ad-supported ecosystem I’ve just described. Silicon Canals uses analytics. We exist inside this system. I’m not writing from a position of purity.
But I think the first step toward changing any system is naming it accurately. And the accurate name for what happens when you tap “I agree” is not “personalized advertising” or “the free internet” or even “surveillance capitalism,” though Shoshana Zuboff’s framework remains the most useful macro-level description. The accurate name, at the level of who wins and who loses, is a class system. It has tiers. It has extraction points. It has a labor force (users), a managerial class (publishers and developers), and a capital-owning elite (platforms and infrastructure firms). It has ideology (“free services,” “personalization,” “connecting the world”) that justifies the extraction to those being extracted from.
The interventions that might actually matter would operate at the structural level. Mandatory revenue transparency in programmatic supply chains, so publishers and users can see where the money goes. Fiduciary duties for data brokers, similar to the obligations financial advisors owe their clients. Collective data rights, modeled on collective bargaining, that would allow groups of users to negotiate terms rather than face the system as atomized individuals. And, most ambitiously, public data trusts that could serve as alternatives to the broker economy, allowing data to be used for research and public benefit without routing it through extraction-optimized private infrastructure.
None of these ideas are original to me. They’ve been proposed by scholars, regulators, and civil society organizations for years. What’s been missing is the political will to implement them, because the companies that would be most affected are also the most effective lobbyists in Brussels, Washington, and increasingly, Singapore and Delhi.
The 50,001st tap
I still tap “I agree.” I did it this morning. The cost of opting out of the modern internet is social and professional isolation, and I’m not willing to pay it. Most people aren’t, and that calculation is itself evidence of the system’s coercive structure: when refusal carries consequences that severe, agreement ceases to be meaningful consent.
What I can do, and what I think more writers and technologists should do, is refuse to describe this system in the euphemistic language it has built for itself. “Data-driven advertising” is a supply chain with human beings as raw material. “Consent management” is the engineering of acquiescence. “The free internet” is a wage arrangement in which the wages are invisible.
The class system hiding in plain sight isn’t hiding because it’s secret. The filings are public. The auction mechanics are documented. The money flows are traceable, if imperfectly. It’s hiding because we’ve been given a vocabulary, a set of tidy metaphors about exchange and choice and personalization, that makes it nearly impossible to see the structure for what it is. Every time you tap that button, you’re participating in one of the largest wealth transfers in human history. The least we can do is call it by its name.
Feature image by Aedrian Salazar on Pexels
