Zuckerberg Wrote in 2018 That Provocative Posts Get More Engagement. New Hampshire’s $224 Million Meta Settlement Leaves the Adult Feed Alone.

Independent New Hampshire Journalism · Northfield, NH

Facebook’s chief executive, the company’s internal documents and a study of 2.7 million posts point the same way: a feed ranked by reaction rewards the post that provokes one. The state’s settlement sets clocks and a chronological option for teenagers. GSR found nothing in it that changes how the feed is ranked for the adults in a town group.

On November 15, 2018, Mark Zuckerberg posted a long note to Facebook about a problem of incentives inside his own product. Left unchecked, he wrote, “people will engage disproportionately with more sensationalist and provocative content.” The closer a post comes to the line of what the company bans, the more people react to it. They do it, he added, even when they say afterward that they did not like what they saw.

Sixteen days earlier, PBS had aired Frontline’s two-part documentary The Facebook Dilemma. Former employees, an early investor and an Egyptian activist described the same machine from the outside. A feed that ranks posts by how many people react will promote the posts that make people react, and hostility does that job well.

That machine is where much of New Hampshire now argues. Forty percent of American adults told the Pew Research Center in August 2026 that they regularly get news on Facebook, more than on any other social site. That same month the state signed a settlement with Meta worth up to $224 million. Its product terms protect teenagers. GSR found no term that changes how the feed is ranked for adults. The company named the problem in 2018, built a brake, and in 2025 said it was easing off. New Hampshire took the money and left that part of the product as Meta built it.

Full disclosure: Granite State Report runs a Facebook page and is building its own verified network for New Hampshire residents, so the editor has a stake in this argument. The facts are checked in the table at the bottom. The opinions are mine.

Related coverage: GSR’s October 19, 2026 piece, on the verified network GSR is building, reports Meta’s ad revenue for all of 2025 and the end of its U.S. fact-checking program. This article uses Meta’s filing for the third quarter of 2026, so the revenue figures in the two pieces cover different periods. Facebook, Free Speech, and the Case for Public Utility Regulation argues the regulatory question, and The Algorithm Is Watching covers bots and moderation.

What the film showed

The Facebook Dilemma aired on October 29 and 30, 2018, with James Jacoby as correspondent. Its account of the News Feed is simple. The feed is a ranked list, and the ranking learns from what you have reacted to. Sandy Parakilas, a platform operations manager at Facebook in 2011 and 2012, told Frontline the feed was designed to keep people scrolling and liking.

Wael Ghonim, whose Facebook page helped start the 2011 protests in Egypt, described what followed. Divisive posts drew more likes, comments and shares, he said, so the system showed them to more people. Roger McNamee, an early Facebook investor, gave the business reason. Fear and anger produced more engagement, which meant more time on the site and more advertising value. In his words, “polarization was the key to the model.”

The Facebook executives Frontline interviewed gave a common answer: the company had been too slow. Naomi Gleit, then vice president of social good, said she regretted that the problem had not been a priority.

What Facebook wrote down

Zuckerberg’s note, titled “A Blueprint for Content Governance and Enforcement,” came with a chart. As TechCrunch described it that day, the line showed engagement climbing as content approached the edge of the rules. His fix was to train the ranking system to push such posts down, in the feed and in recommendations, so the curve would bend the other way.

Internal documents disclosed in 2021 show how the ranking had weighed reactions before that. The Washington Post reported that beginning in 2017, Facebook’s formula counted each emoji reaction, the angry face included, as worth five likes. By 2019 the company’s own data scientists had found that posts drawing angry reactions were disproportionately likely to carry misinformation, toxicity and low-quality news.

The last row is the one to remember. After the angry reaction went to zero, the Post reported from the documents, users saw less misinformation and less graphic violence, and their activity on the site did not drop. A Facebook spokesperson told the Post the company demotes content that draws a disproportionate share of angry reactions.

What outside research measured

The best outside measurement of the raw material comes from a team led by Steve Rathje at the University of Cambridge, published in the Proceedings of the National Academy of Sciences in 2021. The researchers analyzed more than 2.7 million Facebook and Twitter posts from American news outlets and members of Congress. Each word referring to the other political side raised the odds that a post would be shared by about 67 percent.

Posts about political opponents drew more than twice as many angry reactions as posts about a writer’s own side drew hearts. Rathje’s summary: “out-party hate is much better at capturing our attention online than in-party love.”

That study measured what people share. It did not test what the ranking does with it, and the strongest test of the ranking cuts both ways. During the 2020 election, researchers moved consenting Facebook and Instagram users to feeds in plain chronological order for three months and published the results in Science in 2023. Those users spent far less time on the platforms. They saw more political content and more content from untrustworthy sources, and on Facebook they saw less uncivil content. Their polarization and political knowledge did not measurably change.

So the broad claim, that the algorithm by itself makes Americans hate each other, is not proven. The narrower claim holds up. The ranking decides what you see and how long you stay, and a ranking built on reactions favors the posts that provoke them.

The brake, and the foot coming off it

The demotions Zuckerberg promised in 2018 were the brake. On January 7, 2025, Meta’s chief global affairs officer, Joel Kaplan, announced the company was ending third-party fact-checking in the United States, cutting back content demotions and pointing its automated enforcement at illegal and high-severity violations. Political posts from people and pages a user follows would be ranked like any other content, and Meta would recommend more of them. Kaplan called the company’s earlier decision to show people less civic content “a pretty blunt approach.”

The reason is in the company’s books. In the three months that ended September 30, 2026, Meta reported $63.06 billion in advertising revenue out of $64.60 billion in total. That is 97.6 cents of every dollar. Advertising is sold against attention, and McNamee’s arithmetic from 2018 still describes the business: more engagement, more time, more ad value.

What New Hampshire bought

New Hampshire sued Meta in Merrimack County Superior Court in 2023, alleging that Facebook and Instagram were designed to hook young users. In December 2024 the court refused to dismiss the case. According to the Department of Justice, the judge held that neither Section 230 of the federal Communications Decency Act nor the First Amendment shielded Meta, because the claims rested on the company’s own design choices and not on what users posted.

Attorney General John Formella signed a multistate settlement on August 25, 2026, and a proposed consent judgment was filed in federal court in Oakland, California, the next day. New Hampshire’s share is up to $224,344,141 in ten installments. About 70 percent is guaranteed. The rest depends on Snapchat, TikTok and YouTube coming under comparable terms, the state said. A trial had been set for January 2027. In the judgment, “Meta denies the allegations against it” and admits no wrongdoing.

The product terms are written for teenagers, defined as users from 13 through 17. They get a default two-hour daily limit, a notice after 15 minutes of continuous use, an overnight block and muted notifications during school hours. They also get something more interesting: the option to make a feed “displayed in chronological order” their default, with a reminder of that option every 90 days.

The agreement treats a feed without engagement ranking as a protection worth offering a 16-year-old four times a year. It requires no such offer to her parents. Gov. Kelly Ayotte said at the announcement that tech companies had used “their algorithms and their platforms for profit.” The settlement she was announcing reaches those algorithms only for minors.

The adults are the ones arguing about school budgets, warrant articles and zoning in town Facebook groups. They vote. They are also the people Zuckerberg was describing in 2018.

What Concord can ask for

The state has already won the argument that matters. A New Hampshire court has said a design claim against Meta can go forward despite Section 230. Frances Haugen, the former Facebook product manager who testified to the U.S. Senate on October 5, 2021, proposed the same target: leave the law’s protection for what users post and remove it for engagement-based ranking. She favored a return to chronological feeds.

New Hampshire does not need Congress to ask three things. Legislators can ask Meta to give adults here what the settlement gives teenagers, a chronological default that the company reminds them about. They can ask that the independent audits the settlement requires be published. And the state, which decides how its payments are spent, can put part of the money toward measuring what the feed does to civic life in this state, where the evidence is thinnest.

For readers, the Cambridge numbers suggest one test before sharing. If a post is mostly about how awful the other side is, the machine is counting on you.

Granite State Report asked Meta and the New Hampshire Department of Justice for comment by email on November 11, 2026. Neither had replied when this article was published. Any reply will be added here with its date.

Zuckerberg drew the curve himself. A state that just collected up to $224 million over what the product does to teenagers has every reason to ask what it does to everyone else.

— Dexter Dow, Granite State Report

Fact check

Editor’s note. This is a signed analysis by the editor, who discloses above that GSR is building a competing network. Sources: PBS Frontline, Meta’s newsroom and its filing with the Securities and Exchange Commission, the Pew Research Center, the proposed consent judgment and Settlement Agreement, the State of New Hampshire’s announcements, the University of Cambridge, the NYU Center for Social Media and Politics, and reporting by The Washington Post, TechCrunch, MIT Technology Review, NHPR and WMUR. Rows marked Attributed rest on the named source or are the editor’s analysis; the Washington Post, PNAS and Science findings are cited through the named summaries and republication. Corrections: Granite State Report corrects verified errors at the top of the article, states what changed and when, and never removes the note.