The feed is not a window.
It is a machine.

The business model

You are the inventory

Meta generated $200.97 billion in revenue in 2025. Of that, 98 percent came from advertising — not from anything you buy or subscribe to.^[1] The thing being sold is access to you: your attention, your data, and a psychological profile refined by everything you click, pause on, search for, and scroll past.

Advertisers can target you by age, location, job, relationship status, political leanings, what you have bought, what you almost bought, and what time of day you are most impulsive. The average price per ad on Meta rose 9 percent year over year in 2025 — the targeting keeps getting more precise, so access to you keeps getting more expensive.^[1] Every minute you spend scrolling is inventory. That single fact explains almost every design decision that follows.

The machine, step by step

Harvest. Profile. Provoke.

Each step feeds the next. Provoking you generates more behavior to harvest, which sharpens the profile, which makes the provocation more precise. The sections below document each stage with the companies' own experiments, internal presentations, and executives' own words.

Engineered outrage

Anger is the product that sells itself

The steering is not incidental. When Facebook introduced emoji reactions in 2017, the "angry" reaction was weighted five times more than a like in the ranking algorithm — systematically boosting inflammatory content into more feeds.^[3] Internal data scientists warned leadership that this was amplifying the most inflammatory material on the platform. Change came slowly, because reducing inflammatory content also reduced engagement.^[3]

In January 2018 Mark Zuckerberg announced a feed change to prioritize "meaningful social interactions." Internal research found it backfired: divisive posts generate the most comments and reshares, so the change boosted outrage. European political parties told Facebook the new algorithm was forcing them to adopt more negative, extreme messaging just to maintain reach.^[3]

Addiction mechanics

Designed to be hard to put down

The business model needs one more thing from you: time. The platforms use a set of techniques borrowed directly from behavioral psychology — the same reinforcement pattern that makes slot machines the most addictive product in a casino.

Variable rewards. Likes and comments arrive unpredictably. This variable-ratio reinforcement — sometimes a reward, sometimes nothing — is the schedule behavioral psychology identifies as the most habit-forming. Pull-to-refresh mimics a slot machine lever.^[6]

Infinite scroll. The feed never ends, removing the natural stopping cues that let you decide you are done. Its inventor, Aza Raskin, has publicly expressed regret, describing the design of such interfaces this way: "It's as if they took behavioral cocaine and just sprinkled it all over your interface."^[7]

Engineered notifications. Push notifications are not a courtesy. Meta engineers optimize their content and timing to maximize the chance you re-open the app.^[6]

Rabbit holes

The recommendation engine radicalizes

The same engagement logic governs what the platform recommends next. Facebook's collaborative filtering creates self-reinforcing loops of ever more homogeneous — and ever more extreme — content.

A 2020 internal study found that 64 percent of people who joined extremist groups on Facebook did so because the algorithm recommended them. The "Groups You Should Join" feature directed users from mainstream conservative groups toward QAnon and white-supremacist groups.^[5] In 2019, an internal test account was recommended increasingly extreme political content — including conspiracy theories and QAnon groups — within days.^[5]

Facebook has known about the narrowing effect for a decade. A 2015 study authored by Facebook's own researchers, published in Science, acknowledged that the News Feed algorithm reduced exposure to cross-cutting viewpoints by roughly 8 percent.^[9]

The outcomes

What the machine leaves behind

A teen mental-health crisis. Rates of major depression among U.S. teen girls rose approximately 145 percent between 2010 and 2020. Major depressive episodes among all U.S. teens roughly doubled, from about 8 percent in 2010 to over 15 percent by 2019, and self-harm emergency-room visits for girls aged 10–14 nearly tripled between 2009 and 2015.^[10] Meta's own internal research found Instagram makes body image issues worse for one in three teen girls.^[3] The full record is in the Instagram case file.

A misinformation engine. In the final three months of the 2016 U.S. campaign, the top 20 fake news stories on Facebook out-performed the top 20 stories from major legitimate outlets.^[11]

Russian Internet Research Agency content reached an estimated 126 million Americans; health misinformation was viewed an estimated 3.8 billion times in a single year during the pandemic; and the UN found Facebook played a "determining role" in the genocide against the Rohingya in Myanmar.^[12] The details are in the Facebook case file.

And it works in reverse. When a randomized group of users was paid to deactivate Facebook for four weeks in an NYU/Stanford experiment (2,743 participants recruited), their subjective well-being improved by an amount equal to 25–40 percent of the effect of typical clinical interventions for depression, their political polarization decreased, and they kept using Facebook less even after the study ended.^[13] A University of Pennsylvania trial found that merely limiting social media to about 30 minutes a day produced significant reductions in loneliness and depression within three weeks.^[14] Turn the machine off, and the harms recede. See what you get back.