The News Feed and the Trillion-Dollar Timeline

A dozen items from two thousand

Facebook's News Feed launched in September 2006 with a simple premise: instead of making a user visit each friend's profile page to see what had changed, the site would assemble a single, running stream of everything happening across a person's network and show it to them on arrival. The idea drew immediate complaint, much of it about privacy, since it made visible at a glance what had previously required deliberate searching. Facebook kept the feed and spent the next decade refining the harder problem underneath it: with a network large enough, no user could see everything their connections posted, so something had to choose.

That choice is the real invention. By the 2010s, the feed was selecting roughly a dozen items to display from an average pool of about 2,000 potential updates available on a given visit, according to Wikipedia's account of the product. The overwhelming majority of everything posted to a user's network on any day was never shown to them at all. What determined the dozen that were was not chronology, and had not been for years, but a formula assigning each candidate a score and keeping only the highest.

That formula had a name by 2011: EdgeRank, built on three factors. Affinity measured how closely a user interacted with the person or page that posted. Weight assigned more value to certain kinds of content, a photo scoring differently than a plain status update. Time decay reduced a post's score the longer it sat unseen. Multiplied together, the three produced a single number for every candidate post, and the highest-scoring dozen or so won the feed, according to summaries of Facebook's own engineering documentation published by GeeksforGeeks and Martech Zone.

EdgeRank did not last as a fixed formula. Facebook replaced it with a fuller machine-learning ranking system that weighed a far larger set of signals than three variables could hold, and stopped publishing anything resembling a formula a reporter or a researcher could reconstruct by hand. The shift mattered beyond engineering. A three-factor equation could be argued with, audited, and criticized in public. A model trained on billions of engagement events could not, and the accountability problem that created would resurface later, at scale.

EdgeRank's three factors were public knowledge in outline, and marketers spent much of the early 2010s reverse-engineering it in practice: posting at the hours most likely to catch a low time-decay score, favoring photos over plain links because Weight rewarded them, buying comments and shares to inflate a post's apparent Affinity. Facebook's later move to machine-learning ranking closed off much of that gaming. It also closed off the ability of anyone outside the company to state, in three named variables, why one post reached a screen and another did not.

Ranking rules become market capitalization

On September 14, 2012, Facebook crossed one billion monthly active users, a threshold Mark Zuckerberg announced publicly on October 4 of that year, according to TechCrunch's coverage at the time. The number is worth pausing on. It meant the feed's selection process, a dozen items chosen out of roughly 2,000 candidates, was now running for more than a billion people, repeatedly, every day. A ranking decision made inside one product team was operating at a scale no newspaper editor or television scheduler in history had ever reached.

Four months earlier, on May 18, 2012, Facebook had gone public. The IPO valued the company at a peak market capitalization of more than $104 billion, one of the largest technology stock offerings on record, per Wikipedia's account of the listing. Investors buying into that valuation were not pricing a social utility for keeping in touch with friends. They were pricing a ranking engine's demonstrated ability to hold a user's attention long enough, and often enough, to show that user advertising tuned to what had already kept them scrolling.

The bet paid out in a way few 2012 skeptics of the IPO price anticipated. As of 2023, advertising accounted for 97.8 percent of Meta's total revenue, according to Wikipedia's summary of the company's financial disclosures. Almost the entire business now runs on the commodity the feed's ranking produces: attention, measured, sorted, and sold to the highest bidder for a given audience segment. This is the surveillance-advertising model, and while Facebook did not invent behavioral ad targeting on its own, it scaled the model to a size and a precision that made it the template every major platform that followed would adopt in some form.

None of that scale required a sales team calling advertisers one account at a time, the way broadcast and print advertising had for decades before it. The feed's ranking signals doubled as targeting signals: the same affinity and engagement data that decided what a user saw could also describe which of that user's traits, interests, and behaviors an advertiser could buy access to. Selling attention and measuring it had become a single operation inside the same product, which is a large part of why the advertising share of Meta's revenue could climb as high as it has.

The market did not treat this as a one-company story for long. On November 7, 2013, Twitter priced its own IPO at $26 a share, valuing the company at more than $18 billion, and shares opened 73 percent higher on their first day of trading, pushing the valuation past $31 billion before the market closed, according to financial press coverage from Yahoo Finance, TIME, and CNN. Twitter's ranking systems in 2013 were far simpler than Facebook's, still closer to a reverse-chronological timeline than an engagement-optimized feed. That investors extended Facebook-scale multiples to a second social platform within about eighteen months shows how quickly the market had generalized the underlying logic: an engagement-ranked feed, however unfinished its algorithm, was now understood as an advertising business with a growth curve worth a great deal more than the sum of the posts running through it.

The bubble the ranking built

The same year EdgeRank formalized Facebook's ranking logic, 2011, the writer and activist Eli Pariser published a book naming the pattern he saw forming across the industry. The Filter Bubble argued that algorithmic personalization, tuned to show each user more of what its model expected them to engage with, tends over time to narrow the range of information a person sees rather than broaden it. The term stuck, and it became one of the standard critiques leveled at every ranked feed built after it, Facebook's included.

Pariser's argument was a hypothesis in 2011, built on the limited public description Facebook had given of how EdgeRank worked rather than on access to the ranking system itself. That did not stop it from becoming one of the most repeated frames in the decade of platform criticism that followed, cited well beyond Facebook, in arguments about search results, video recommendations, and eventually the feeds of every company that adopted engagement ranking in the years after 2011.

That critique cuts both ways. The same ranking engine that could surface a friend's vacation photograph could just as readily surface an eyewitness video from a protest a traditional broadcaster had not covered, or a small local business's post reaching an audience no newspaper classified section had ever offered it. Distribution that once required an editor's assignment, a broadcaster's license, or a printer's budget now required only a high enough affinity, weight, and time-decay score. That is a genuine liberation of who could reach an audience.

It came bundled with the concentration Pariser described. Because affinity and weight rewarded continued interaction, and because time decay punished anything that failed to draw attention quickly, the formula had no built-in preference for accuracy, importance, or civic value; it optimized for what a user's own past behavior had taught the model they would click, comment on, or linger over. A user increasingly saw the version of the network their prior engagement had trained the ranking system to expect they wanted, decided by a formula they could not see and had never agreed to in any specific sense. Liberation and concentration were not two separate outcomes of the feed. They were the same mechanism, read from two directions.

Rewriting the formula after an election

In January 2018, Facebook announced a significant overhaul of the News Feed, one it described in its own newsroom statement as prioritizing "meaningful social interactions": comments, reshares, and other active engagement, ranked above passive content a user might scroll past without responding to. Facebook framed the change as a response to internal research suggesting passive consumption correlated with lower reported wellbeing among users, a quality-of-life argument rather than a political one.

The timing told a second story. The overhaul followed roughly a year and a half of American congressional hearings, press investigation, and public scrutiny of Facebook's role in the 2016 United States presidential election and the Cambridge Analytica data-harvesting scandal that broke into public view in early 2017. The company's own account of the 2018 change explicitly linked it to that period of election-related scrutiny, and it is difficult to read the reweighting as unconnected to a controversy that had put the feed's ranking rules in front of a legislature for the first time.

What the change actually did to the content mix a user saw is harder to state with the same confidence as the fact of the announcement. Facebook described its intent in qualitative terms and did not publish the underlying formula, so outside estimates of the reweighting's real effect are inferences drawn from later reporting rather than a measured audit of the algorithm itself. The company argued that rewarding comments and reshares over passive viewing would produce a healthier feed. Critics countered that comments and reshares are exactly the reaction emotionally charged, divisive, or outrage-driven posts tend to generate in higher volume than measured, accurate ones, meaning the same formula meant to reduce low-value content could just as easily have rewarded the most polarizing content in a user's network. Both readings are plausible from the public record. Neither is settled by it.

Publishers whose businesses had grown around the feed's earlier traffic patterns felt the reweighting immediately. When Facebook shifted its ranking weight toward comments and reshares between friends and family, and away from passive content, a smaller share of what the feed served after January 2018 pointed a user away from the platform toward a news article at all. A single change to a private ranking formula, made for reasons the company described as internal and civic, reshaped a share of the referral traffic an entire industry had come to depend on.

Whoever ranks the feed governs the room

By the time of the 2018 overhaul, Facebook's feed had grown well past the scale of any single national broadcaster. With more than two billion monthly users, Wikipedia's entry on the News Feed describes it as having become "the most viewed and most influential aspect of the news industry," a claim that would have sounded implausible for any single company to make about itself a decade earlier, when broadcast licenses and newspaper circulation still defined who could reach a national audience.

That scale changed what a ranking decision was. When a feed becomes the primary channel through which a meaningful share of a country's population receives its news, the formula that decides what surfaces in that feed stops functioning as a private product choice and becomes, in effect, a national information policy, set by a product team's internal metrics rather than by a legislature's broadcast rules, a newsroom's editorial standards, or a printer's licensing terms. No election put that formula in place, and no election could remove it.

The January 2018 rewrite is itself evidence of this shift. A company does not typically redesign its central ranking system in response to a competitor's move or a quarterly earnings miss and then explain the redesign publicly by referencing a national election and a data-privacy scandal that had reached a legislature. Facebook did exactly that, and in doing so acknowledged, whether or not it used the phrase, that its ranking rules had become a matter states had a legitimate interest in.

Broadcast history had at least built formal channels for this kind of power: licenses granted and revoked by regulators, rules on political airtime, ownership limits meant to stop any single network from reaching too much of a country alone. None of that architecture existed for a ranking formula that changed on its own schedule, inside one company, without the public rulemaking that had governed the media it was now displacing as the country's primary news channel.

Here too the record cuts two ways. Supporters of ranked, algorithmic distribution note that it broke the near-monopoly a small number of broadcasters and publishers once held over what a national population saw each day, spreading that power more widely than the pre-feed era ever managed. Critics note that spreading the power of distribution did not spread the power of ranking; one company's internal formula change could now reshape what a large share of a country read overnight, an authority no single newspaper editor or television network president had held alone even at the height of the broadcast era. Both claims are defensible. They describe the same feed from opposite ends of the same argument.

The feed's afterlife in the answer engine

The specific mechanism this history describes, an engagement-tuned social feed ranking posts from a user's own network, belongs to its decade. Facebook has kept revising it since 2018, and the platforms that copied the surveillance-advertising model built variations of their own. But the underlying lesson from EdgeRank through the 2018 rewrite does not belong only to that decade: whoever writes the ranking rule for the surface where a population goes looking for information holds real economic power and, past a certain scale, real civic power over what that population believes it knows.

A related discipline is now forming one layer up from the feed. Where the News Feed chose a dozen items out of roughly 2,000 candidates, an AI answer engine chooses one name, one source, or one fact to state in response to a question, out of everything it could have surfaced instead. The mechanism is not the same as EdgeRank's affinity, weight, and time decay, and it would overstate the case to call it a direct descendant. But the discipline the feed's decade first proved out at scale, that a private ranking choice made at the surface where people already look becomes an economic and civic fact whether or not anyone outside the company that made it had a vote, is the same discipline now being tested at the point where an answer, rather than a ranked list, is the whole of what a person sees.

That is also why this history belongs to a study of trust and not only to a study of technology. The feed did not just change what people read; it changed who could be found, cited, and believed by the systems standing between a business and its audience, first inside one company's ranking product, and now, increasingly, inside the model an assistant consults before it answers a question at all.