In 2000, classified advertising generated roughly $20 billion a year for American newspapers. By 2012, that figure was $4.6 billion — a 77 percent collapse in twelve years. The company most responsible for that collapse, Craigslist, had fewer than fifty employees.
Fewer than fifty people, fifteen billion dollars in destroyed revenue. Not through espionage or predatory pricing or regulatory capture — through a free listings page that worked better than the one in the newspaper. Craigslist didn’t steal anything. It did what markets are supposed to do: it offered a superior product at a lower price. The price happened to be zero.
The received version of this story is about journalism. Newspapers died, the internet killed them, and now we’re all worse off. That version is mostly wrong — not in its conclusions, but in its diagnosis. The newspaper industry didn’t collapse because people stopped wanting information. It collapsed because the economic structure that had sustained it for a century was built on a form of artificial scarcity that the internet made irrelevant overnight. What happened next wasn’t a tragedy of carelessness. It was an economic relocation — and understanding where the value went, and why, is what makes it possible to understand what the attention economy actually is.
What newspapers were actually selling
Here’s what a newspaper was, economically, before the internet: a bundle. Classified ads — jobs, apartments, used cars — sat next to display advertising, local sports results, stock prices, weather forecasts, and, somewhere in the middle of all that, journalism. Each component subsidized the others, but the money flowed in one direction. Classifieds were the engine. At their peak, they represented approximately 30 percent of total newspaper revenue. That was the margin that paid for reporters.
What made this bundle work was not the quality of the content. It was the cost of the infrastructure. Printing presses, delivery trucks, carrier networks, regional distribution agreements — building a newspaper required enormous capital investment. That capital investment was the barrier to entry, and the barrier to entry was the source of value. A classified advertiser in Dayton, Ohio, in 1995 had exactly one efficient way to reach people looking for a used Honda Civic in Dayton, Ohio: the Dayton Daily News. Not because the Daily News was the best possible vehicle for that transaction. Because it was the only one.
Economists have a term for this. Newspapers were extracting rent — not from the quality of their product, but from their control of the distribution channel. The reporting, the investigative journalism, the local political coverage that held county commissioners accountable — all of it was cross-subsidized by the rent that flowed from classifieds and display ads. The journalism was real. The business model that funded it was a tollbooth.
The internet didn’t build a better newspaper. It eliminated the tollbooth. When distribution costs went to zero, the capital barrier that had been generating newspaper rent evaporated. Craigslist could offer the same classified marketplace — actually, a better one, searchable and free — without owning a single printing press. The research confirms the scale of destruction: Robert Seamans and Feng Zhu, writing in Management Science in 2014, documented that Craigslist’s entry into local markets between 2000 and 2007 cost the newspaper industry approximately $5 billion in classified revenue.
Television worked the same way, just with a different tollbooth. Broadcast networks controlled spectrum access, allocated by FCC licensing — scarcity imposed by regulation rather than capital cost. Cable multiplied the available channels; the internet finished the job. In both cases, the business model was identical: sell access to a captive audience. In both cases, the internet took the captive part away.
The political cost of a cross-subsidy collapse The classified revenue that vanished didn't fund everything equally. It funded, disproportionately, the journalism that generated the least direct revenue: local political reporting. When Craigslist entered a local market, newspapers didn't cut sports coverage or entertainment listings. They cut political reporters — the most expensive staff producing the least commercially viable content. Djourelova, Durante, and Martin, in a 2025 study published in the Review of Economic Studies, traced the consequences at the county level: Craigslist's entry reduced coverage of local congressional races by an average of 12 percent and measurably increased partisan voting behavior. The economic collapse of local news wasn't just a cultural loss. It was a documented mechanism of political atomization.
The arithmetic of free
The value of controlled distribution didn’t vanish. It relocated. Understanding where it went requires a piece of economic logic that Herbert Simon articulated in 1971, decades before anyone had heard of Google: “A wealth of information creates a poverty of attention.” When information is abundant and free, the scarce resource isn’t the information. It’s the attention required to process it.
Simon wasn’t writing about the internet — he was writing about organizational design. But the principle is a general one, and the internet made it operational at a scale Simon couldn’t have imagined. When every newspaper, every broadcaster, every individual with a blog or a camera phone can distribute content at zero marginal cost, content stops being scarce. Attention — the fixed quantity of human consciousness available to process all that content — becomes the bottleneck. And whoever can organize that bottleneck at scale captures the economic value that used to attach to the content itself.
Google’s business model is the purest expression of this logic. Search is free. The product isn’t search. The product is you — specifically, the organized delivery of your attention to an advertiser at the exact moment you’re expressing commercial intent. When you type “best running shoes” into Google, you’re not Google’s customer. You’re the commodity being sorted and sold. The advertiser pays for access to your attention at a moment of peak receptivity. That intent signal is extraordinarily valuable, and Google holds it at a scale no one else can match.
Facebook’s model is structurally similar but different in texture. Google captures intent-driven attention — people looking for something. Facebook captures ambient attention — people socializing, scrolling, killing time.
Google’s advertising revenue in 2024 was $264.6 billion. Meta’s was approximately $160 billion. The entire American newspaper industry at its 2005 peak — the high-water mark — generated roughly $49 billion in advertising revenue. By 2020, combined advertising and circulation revenue had fallen below $20 billion. The money didn’t disappear. It moved upstream, to the companies that organized the attention newspapers had once captured by default.
The specific wreckage
The Medill School’s State of Local News report, published in 2025 by Northwestern University’s Local News Initiative, puts numbers on what the relocation left behind. Nearly 3,500 American newspapers have closed since 2005 — close to 40 percent of all local papers. Over 270,000 newspaper jobs have been eliminated. Fifty million Americans live in areas the report classifies as “news deserts,” with limited or no access to local news. Two hundred thirteen counties have no news outlet of any kind. Another 1,524 have exactly one. Almost 80 percent of news deserts are in counties the USDA classifies as predominantly rural.
And then there’s the Youngstown Vindicator. A hundred and fifty years of continuous publication — coverage of the KKK, the mafia, local political corruption — a paper built on investigative accountability. It closed on August 31, 2019. A hundred and forty-four employees lost their jobs.
The New York Times, meanwhile, ended 2024 with more than 10.8 million digital subscribers, and by the third quarter of 2025, its total subscriber base had reached 12.3 million. A handful of other national titles built subscription businesses in the millions — the Wall Street Journal among them. These institutions didn’t die. They thrived.
But the survival of elite national outlets doesn’t contradict the structural argument. It confirms it. The Times escaped the cross-subsidy trap not by replicating the platform model — not by selling attention to advertisers — but by building direct reader revenue at a global scale that geography makes impossible for local papers. That’s a structurally different business model: readers pay for the content. No one is selling their attention to a third party. But this exit only works if your journalism appeals to readers worldwide. Readers in London and Seoul and São Paulo choose to pay for Times coverage of the White House and the Middle East. Mahoning County, Ohio, cannot build a subscriber base in Mumbai. The exit from the platform economy requires either becoming a platform yourself or becoming a globally scaled subscription product. The Vindicator could do neither. Almost no local paper can. The mechanism that killed it is intact. The Times found a door that exists for perhaps a dozen institutions on Earth.
The nonprofit alternative and its limits ProPublica, The Texas Tribune, and a growing network of local nonprofit news organizations represent a genuine response to local journalism's collapse. But they cannot replicate its scale. The nonprofit news sector collectively employs several thousand journalists. The newspaper industry lost more than 270,000 jobs. Philanthropy fills a niche — an important one. It doesn't restructure a market. The question isn't whether nonprofit journalism is valuable. It's whether donor-funded reporting can replace an economic model that once sustained tens of thousands of newsrooms through self-sustaining revenue. The honest answer is no.
What the algorithm actually optimizes
The same economic logic that relocated value from content producers to attention aggregators also determines what those aggregators do with the attention they’ve captured.
When a platform’s revenue depends entirely on how much time users spend on it, the platform optimizes to maximize that time. Engagement — likes, shares, comments, continued scrolling — is the proxy metric. The question the algorithm is always answering is: what keeps this person on the platform for another thirty seconds? And the empirical answer, documented by the platforms’ own researchers, is: content that provokes strong emotional reactions. Outrage, fear, moral disgust. These generate the most reliable engagement signals. They travel faster, get shared more, produce comment threads that pull other users in.
The claim here is not that Mark Zuckerberg wanted to make people angry. The claim is that Facebook built a system designed to maximize engagement, and the output of that system was the systematic amplification of divisive content — because divisive content is what engagement-maximization produces. Facebook’s own researchers knew this. Frances Haugen’s 2021 disclosure of internal company documents showed that Facebook’s teams had documented the outrage-amplification dynamic, proposed mitigations, and been overruled by senior leadership because the mitigations would have reduced engagement metrics. They knew. They chose revenue.
William Brady and Molly Crockett, in a 2021 study published in Science Advances, demonstrated the mechanism from the user side. Across two preregistered observational studies on Twitter — 7,331 users, 12.7 million tweets — and two behavioral experiments, they found that users who received more likes and retweets when expressing moral outrage were more likely to express outrage in subsequent posts. The platform’s reward structure was training users to be angrier. Not because anger was what users wanted. Because anger was what the feedback loop reinforced.
This isn’t about one company. Any platform whose revenue depends on maximizing attention time through a free product faces identical incentives. Measure what holds attention, optimize for it, let the feedback loop run. Change the management. Change the country of origin. The incentive is the same. The system produces what it is optimized to produce.
What Facebook knew, and when The Wall Street Journal's "Facebook Files" investigation, published in September and October 2021, drew on internal documents showing that Facebook's researchers had repeatedly flagged the outrage-amplification problem. A 2018 change to the News Feed algorithm, designed to prioritize "meaningful social interactions," had the predictable effect of amplifying emotionally provocative content, because that's what generates interactions. Internal presentations warned senior leadership. Proposals to weight the algorithm toward less inflammatory content were rejected on the grounds that they would reduce engagement. Haugen's Senate testimony on October 5, 2021, put those documents into the public record. The company's response was to rename itself Meta.
Why markets do this
Google has held more than 89 percent of the global search market for over a decade. Not for lack of competition — Microsoft has spent billions on Bing, and every few years a new search startup arrives with venture funding and a pitch about doing search differently. None of them have made a dent. Facebook has weathered the Cambridge Analytica scandal, congressional hearings, advertiser boycotts, a whistleblower, and a rebrand. Its core advertising demographics barely moved.
Why can’t anyone compete with these companies? The answer isn’t that they’re smarter or more ruthless, though they may be both. The answer is structural.
Network effects: the value of a platform to any individual user increases with the number of other users. More users produce more content, which improves recommendations, which attracts more advertisers, which funds better infrastructure, which attracts more users. Carl Shapiro and Hal Varian identified this dynamic in their 1999 book Information Rules: information goods markets tend toward tipping because of positive feedback loops. Once a platform reaches critical mass, the gap between it and competitors becomes self-reinforcing. The loop runs in one direction. And the economics of digital goods compound it — serving one additional user of Google Search costs Google essentially nothing. A newspaper that doubled its circulation doubled its printing and distribution costs. A platform that doubles its user base barely notices. Once fixed infrastructure costs are covered, additional revenue is nearly pure profit, which funds the algorithmic improvements that widen the gap further.
You cannot compete with Google on price. Google is free. You cannot compete with Facebook on price. Facebook is free. That’s not a business strategy — it’s a structural barrier created by two-sided markets. Platforms serve two distinct customer groups — users and advertisers — connected through the platform. Users get a free product funded by advertising. Advertisers get access to a user base they can’t efficiently reach elsewhere. A would-be competitor has to offer its product for free too, which means it also needs advertising revenue, which means it needs a user base large enough to attract advertisers, which means it’s caught in the same chicken-and-egg loop the incumbents already solved a decade ago. Competition happens on recommendation quality and engagement features, where the incumbent with more data has a structural advantage that grows with every user added.
Google and Meta combined held approximately 48.7 percent of all US digital advertising revenue in 2023, down from a peak of 54.7 percent in 2017. But that decline doesn’t mean the market is becoming less concentrated. It means the structural logic is replicating. Amazon built a $30-billion-plus advertising business by capturing product-search intent — the same mechanism Google uses for web search, applied to shopping. TikTok captured entertainment attention using the same engagement-optimization playbook. Every new platform that organizes attention at scale becomes another extraction point. The mechanism doesn’t weaken. It propagates.
The limits of what regulation can do
If the concentration is structural and the incentives are determined by the business model, then whether regulation can fix the problem is an empirical question, not a moral one. And the empirical record is not encouraging.
The Australian News Media Bargaining Code, passed in March 2021, was the most ambitious attempt to redistribute value from platforms to publishers. It worked — for a while. More than thirty commercial agreements were struck between Australian news organizations and platforms. Newsrooms hired journalists with the resulting funds. The Treasury’s December 2022 statutory review called it a success.
Then, on February 29, 2024, Meta announced it would not renew any of its deals with Australian media organizations and would shut off Facebook News in Australia entirely. Cost to Australian news organizations: approximately A$70 million annually. As of that date, no digital platform corporation had been designated under the Code’s enforcement mechanism. The Code’s stick had never been used, and the carrot had just walked away.
What the Code did: temporarily redistribute some advertising-adjacent revenue. What it didn’t do: change what platforms optimize for, who they sell attention to, or how their algorithms decide what content gets amplified. It rearranged some money without touching the machine.
The US Journalism Competition and Preservation Act would have allowed news publishers to collectively negotiate with platforms — an antitrust exemption for collective bargaining. It failed to pass the Senate. Even had it passed, it faced the same structural limitation: revenue reallocation without incentive alteration.
The EU Digital Services Act, effective February 2024, is more structurally ambitious. It requires Very Large Online Platforms to conduct annual systemic risk assessments, including how their recommender systems contribute to disinformation amplification. It mandates mitigation measures. But the assessments are largely self-reported, enforcement has been slow, and the structural incentive to maximize engagement hasn’t changed — only the regulatory requirement to write a report about it.
What would actually change the incentives? The business model would have to change. Three architectures are conceivable. First: subscription at platform scale. If users paid directly for search and social media, platforms wouldn’t need to maximize advertiser access to attention time. But every experiment with platform subscriptions — X Premium, Meta Verified — has been an add-on to the advertising model, not a replacement for it. No platform has voluntarily given up its primary revenue source.
Second: mandated algorithmic intervention. Require platforms to adopt recommendation systems that don’t optimize for engagement time. Possible in principle. In practice, extraordinarily difficult to define in law — too vague and it’s unenforceable, too specific and it can’t adapt — and likely to push users toward less-regulated alternatives.
Third: structural separation. Require platforms to be either content recommenders or advertising businesses, not both. Break the link between what gets amplified and what generates advertising revenue. This would be the most structurally effective approach. It is also the most politically impossible.
The public subsidy alternative There is an approach that doesn't try to fix the platform economy but routes around it entirely: direct public funding for journalism. The BBC model. Government support for local news. This is conceptually distinct from regulating platforms — it accepts the attention economy as a given and tries to sustain journalism outside it. But in the American context, the obstacles are severe: political capture risk is real (which party decides who gets funded?), the scale gap between what philanthropy and subsidy can provide and what the market once sustained is enormous, and no political constituency is organized around the idea of paying taxes so that county commissioners get covered.
The honest conclusion is this: current regulatory approaches cannot reach the structural problem. They redistribute some revenue or mandate some self-assessment. They don’t change what platforms are economically rewarded to do. And the difficulty isn’t only a failure of political will. The business model that produces outrage amplification and news-desert creation is the same business model that underpins the cheapest, most efficient information distribution system ever built. Dismantling the incentive structure means paying more for distribution, or accepting less of it.
No political constituency is organized around that trade-off. And it isn’t clear one could be — because organizing a political constituency now requires competing for attention on the very platforms whose incentive structure you’re trying to change. The regulators have Instagram accounts. The reform campaigns optimize for engagement. The members of Congress who hold hearings on algorithmic amplification announce those hearings on Twitter.
Twenty billion dollars in classified advertising revenue became four point six billion. Forty-some employees at a free listings site erased a subsidy that had sustained local journalism for a century. That wasn’t destruction. It was relocation — value moving from those who produced content to those who organized the reading of it. Everything that followed, from the 3,500 closed newspapers to the algorithmic amplification of outrage to the regulatory interventions that couldn’t reach the underlying mechanism, is what markets do when the product is free and the commodity being sold is your attention.
The question isn’t whether someone should do something. The question is whether the institutions built to govern this economy can do so from inside a system that has already captured the resource they’d need to organize the effort: the public’s attention.
Gen AI Disclaimer
Some contents of this page were generated and/or edited with the help of a Generative AI.
Media
Key Sources and References
Seamans, Robert and Feng Zhu. “Responses to Entry in Multi-Sided Markets: The Impact of Craigslist on Local Newspapers.” Management Science, vol. 60, no. 2, February 2014, pp. 476–493. DOI: 10.1287/mnsc.2013.1785.
Djourelova, Milena, Ruben Durante, and Gregory J. Martin. “The Impact of Online Competition on Local Newspapers: Evidence from the Introduction of Craigslist.” Review of Economic Studies, vol. 92, no. 3, May 2025, pp. 1738–1772. DOI: 10.1093/restud/rdae049.
Simon, Herbert A. “Designing Organizations for an Information-Rich World.” In Greenberger, Martin (ed.), Computers, Communications, and the Public Interest. Johns Hopkins Press, 1971.
Northwestern University Local News Initiative. “The State of Local News 2025.” Medill School of Journalism, Media, Integrated Marketing Communications, October 2025. https://localnewsinitiative.northwestern.edu/projects/state-of-local-news/2025/
Brady, William J. and Molly J. Crockett. “How Social Learning Amplifies Moral Outrage Expression in Online Social Networks.” Science Advances, vol. 7, no. 33, August 2021. DOI: 10.1126/sciadv.abe5641.
Haugen, Frances. Written testimony before the United States Senate Committee on Commerce, Science, and Transportation, Subcommittee on Consumer Protection, Product Safety, and Data Security, October 5, 2021.
Horwitz, Jeff, et al. “The Facebook Files.” Wall Street Journal, September–October 2021.
Shapiro, Carl and Hal R. Varian. Information Rules: A Strategic Guide to the Network Economy. Harvard Business School Press, 1999.
Alphabet Inc. Fourth Quarter and Fiscal Year 2024 Results. SEC filing, February 2025. https://www.sec.gov/Archives/edgar/data/1652044/000165204425000010/googexhibit991q42024.htm
Meta Platforms Inc. Fourth Quarter and Full Year 2024 Results. February 2025. https://investor.atmeta.com/investor-news/press-release-details/2025/Meta-Reports-Fourth-Quarter-and-Full-Year-2024-Results/
Australian Government Treasury. News Media and Digital Platforms Mandatory Bargaining Code: Statutory Review. December 2022.
Norton Rose Fulbright. “Australia’s News Media Bargaining Code: What Happens Next.” March 2024.
European Union. Digital Services Act, Article 34: Risk Assessment. Regulation (EU) 2022/2065, effective February 17, 2024.
The Vindicator (Youngstown, Ohio). “The End of Our Story: The Vindicator’s 150-Year Run Ends Today.” August 31, 2019.
StatCounter Global Stats. Search Engine Market Share Worldwide. https://gs.statcounter.com/search-engine-market-share
The New York Times Company. Fourth Quarter and Full Year 2024 Results. February 2025; Third Quarter 2025 Results. November 2025.
Willens, Max. “Meta and Google’s hold on digital advertising loosens as TikTok and others gain share.” EMARKETER, June 27, 2022. https://www.emarketer.com/content/meta-google-s-hold-on-digital-advertising-loosens-tiktok-others-gain-share
Craigslist. “About Jim Buckmaster.” craigslist.org. https://www.craigslist.org/about/jim_buckmaster
Pew Research Center. “Estimated advertising and circulation revenue of the newspaper industry.” Journalism Project / State of the News Media. https://www.pewresearch.org/chart/sotnm-newspapers-newspaper-industry-estimated-advertising-and-circulation-revenue/
English 
Español 
Français 
Deutsch 
日本語 
Polski