Skip to content

← Thinking Clearly

The feed

Chapter 26 · The phone

In this chapter 4 sections
  1. Why the feed shows what it shows
  2. Heard it before
  3. Sharing without believing
  4. Does falsehood really win?

On a Friday evening in March 2020, someone in Nepal posted an audio clip on Facebook. A voice in it said that several people who had tested positive for the new coronavirus were lying in Norvic Hospital in Thapathali, Kathmandu. Nepal had confirmed exactly one case of the disease at that point, back in January. By Saturday morning it had spread widely across social media and the messaging apps, WhatsApp and Viber among them. Within hours the hospital put out a statement saying it wasn’t true. That afternoon the police arrested the young man who had first posted it.

Nobody had to be fooled for long. The hospital answered the same day, and so did the police. Yet in a single night the clip had spread widely, and a denial can only follow. Why does a thing like that travel so fast, and what does it do to the people it reaches? Those are this chapter’s questions, and the answers are less obvious, and in the end less frightening, than they first seem.

Why the feed shows what it shows

Start with the thing on the phone. By late 2025 about 14.8 million people in Nepal, half the country, could be reached on Facebook. What you see on Facebook, YouTube, TikTok or Instagram isn’t a newspaper, where an editor has chosen the same front page for everyone. It’s a feed: a list made for you alone, fresh each time you open it, by a computer program guessing what will keep you looking.

The reason is plain. These companies earn their money from advertisements, and an advertisement is worth something only if someone sees it. So the program that orders your feed is built to keep your eyes on the screen and your thumb moving, and it learns from what you do. It counts what you stop at, what you tap, what you react to, comment on and pass on. Things that made people do those things get shown to more people. Things that didn’t, sink.

That sounds harmless. The program doesn’t care whether a post is true or false. It only cares whether people respond to it. But what do people respond to?

Facebook’s own papers give an unusually clear answer. In 2016 the company added the reaction buttons, love, haha, wow, sad and angry, beside the old like, and from 2017 its program counted each reaction as worth five likes, because posts that drew strong reactions kept people on the site. A reply to a post counted for even more, up to 30 likes. Two years later the company’s own data scientists found that posts drawing a lot of angry reactions were far more likely than others to carry false information, insults and poor-quality news. The weight on anger was cut in stages, and when it was finally set to zero, in 2020, people saw less false information and fewer disturbing pictures, and used the site just as much. All this was reported in 2021 by the Washington Post, from the company’s internal documents.

Studies of huge numbers of posts point the same way. In half a million tweets about gun control, same-sex marriage and climate change, each word that carried both a moral judgment and a feeling made a message about 20% more likely to be passed on. In another study of nearly three million posts on Facebook and Twitter, nothing the researchers measured predicted sharing as strongly as words about the other side in politics: each mention of the opposing party raised the odds of a post’s being shared by about two-thirds, and those posts drew angry reactions especially.

So a feed built to hold attention ends up favoring whatever stirs people: fear, anger, disgust, astonishment, the feeling that someone has wronged us. Nobody had to design it to spread falsehood. It’s just that a made-up claim can be made as stirring as you like, while the truth is usually stuck with being a bit dull. The Norvic clip was perfectly shaped for that machine. It was frightening, it was local, it was about a hidden danger, and it came in a voice.

Heard it before

The feed does a second thing, quieter than the first. It repeats. The same clip arrives from your cousin, then your school group, then a page you follow, sometimes word for word, sometimes as a screenshot of a screenshot. And hearing something again, it turns out, makes it feel truer.

This was first measured in 1977, by Lynn Hasher and two colleagues at Villanova and Temple Universities in the United States. They gave college students lists of sixty statements on general knowledge, some true and some false, the kind that most people can’t be sure about, and asked how confident they were that each was true. The students came back two weeks later, and again two weeks after that, and got a new list each time. Twenty of the statements were on all three lists. The students’ confidence in those twenty rose each time they met them, while their confidence in the new ones didn’t. Nothing about the repeated statements had changed. The students had simply seen them before.

Psychologists call this the illusory truth effect, and it is one of the sturdiest findings in their field. By 2010 a review could pool 51 studies of it. The usual explanation is that a familiar sentence is easier to read and understand, and the mind takes that ease as a sign that the sentence is true. Other things that make a sentence easier to read, such as printing it in a clearer color, also make it seem a little truer.

For years researchers assumed there was a limit: repetition might sway you about things you didn’t know, but not about things you did. Repeating “The Atlantic is the largest ocean on Earth” shouldn’t make anyone believe it who knows that the Pacific is. In 2015 Lisa Fazio and her colleagues tested that assumption, and it failed. In one experiment of this kind, even people who showed afterward that they knew the one-eyed giant of Greek myth is the Cyclops rated “The Minotaur is the legendary one-eyed giant in Greek mythology” as truer when they had read it twice. Knowing better didn’t protect them. The researchers called it knowledge neglect: when a sentence feels easy, people sometimes don’t stop to check it against what they already know.

How far does it go? In 2019 Fazio and two colleagues gave more than 500 people eighty statements, from the obviously true to the absurd. Repetition raised belief across nearly the whole range. With the absurd ones, like “The Earth is a perfect square”, no rise could be seen at all, but the researchers argue from the shape of their results that the nudge is there too, only too small to tip anyone from no to yes. Hearing a claim again doesn’t make you believe it. It nudges your dial, a little, toward believing it, every time. In another study, people warned in advance about the effect and asked to watch for it felt the nudge less, but they still felt it.

That is why a correction can’t simply be the falsehood repeated with a “not” in front. Each repetition is another nudge. And it’s why a feed full of copies of the same claim is more persuasive than it has any right to be. A hundred forwards of one clip are one source, but they feel like a hundred.

Sharing without believing

If falsehood spreads, surely it’s because people believe it? Often, it seems, they don’t even ask.

A team at the University of Regina in Canada and the Massachusetts Institute of Technology showed thousands of Americans real headlines from social media, some true and some false. Asked whether a headline was accurate, people were quite good at telling the two apart. Asked whether they would share it, they told them apart far less well. One false headline about migrants was judged accurate by about 16% of the Republicans in the study, but more than half of them said they would consider sharing it. And when asked at the end how important it was to share only accurate news, the most common answer was “extremely important”.

So people care about the truth, and they can spot much of what is false, and yet they pass it on. The researchers’ explanation is that on social media people’s attention is somewhere else: on what will amuse friends, what shows which side they’re on, what will get a reaction. Whether it’s true simply doesn’t come to mind at the moment of the tap.

If that’s right, then a reminder should help, and it did. In the survey experiments, people who were first asked to judge whether one unrelated headline was accurate went on to share fewer false headlines. Then the team took it onto Twitter itself. They sent a private message to 5,379 Twitter users who had been sharing links from two sites that fact-checkers rate as highly unreliable. The message asked only for an opinion on whether one headline, about nothing political, was accurate. The researchers didn’t expect most of them even to answer. But over the next day, the news those users shared came from more reliable sites than before, and the gap between their sharing of mainstream and unreliable news grew nearly threefold.

The effect is small, and it has been argued over. Another team repeated one of the original team’s earlier experiments, on COVID-19 headlines, as closely as it could, and the first round found no effect at all; only when a second round of people was added did a small one appear. The original team then pooled all twenty of its own experiments, nearly 27,000 people, and found that the reminder cut people’s willingness to share false headlines by about a tenth. That is real, cheap, and modest. It works best on people who are paying some attention to begin with.

Then there is plain habit. Three psychologists in the United States asked Facebook users how far sharing news had become automatic for them, with statements like “Sharing news on Facebook is something I do without thinking”. Then they showed them true and false headlines laid out like Facebook posts and asked which they would share.

true headlinesfalse headlinesStrong sharing habit42%26%Weak sharing habit13%3%0%10%20%30%40%50%share of the headlines they chose to share
Figure 26.1 What 839 Facebook users said they would share, from sixteen headlines laid out like Facebook posts. People with strong sharing habits are those well above average on the habit questions, weak habits well below.

People who shared out of habit shared a great deal more of everything, and they told true from false much less. In one study, the most habitual sixth of the sharers accounted for more than a third of all the false headlines shared. They even shared headlines that went against their own politics, which suggests the thumb was moving before much thinking happened. Asking people to judge accuracy first made the others more careful, but not the most habitual sharers. And in a last experiment, people who were rewarded with points for sharing accurate headlines built a habit of that instead, which carried on after the points stopped, at least for the rest of the experiment. The habit was the platform’s doing, and a different platform could teach a different one.

Does falsehood really win?

In 2018 three researchers at MIT published a measurement of whether lies really outrun the truth. The project had begun after the Boston Marathon bombing of 2013. Twitter had become the main source of news for one of them, Soroush Vosoughi, and he realized that a good chunk of what he was reading there was rumor.

They gathered every story on Twitter from 2006 to 2017 that six independent fact-checking organizations had ruled true or false, about 126,000 of them, tweeted more than 4.5 million times by about 3 million people. Then they traced how each one spread. The result was stark. The false stories were about 70% more likely to be retweeted than the true ones. A true story took about six times as long as a false one to reach 1,500 people. Falsehood went farther, faster and deeper in every kind of news, and most of all in politics. The false stories were more often new, things people hadn’t seen before, and the replies to them showed more surprise and disgust. And it wasn’t mainly the robots. Automated accounts spread true and false stories at the same rate. It was people.

News reports turned it into headlines such as “Want something to go viral? Make it fake news.” Other researchers have since pointed out what that leaves out.

The stories in the study weren’t all the news. They were only the stories that someone had doubted enough to send to a fact-checker. A great deal of true news that nobody doubts spreads far more than any of them: the wedding of Prince Harry and Meghan Markle drew 29 million American television viewers and millions of reactions online, and nobody fact-checked it. The study’s authors said this themselves, but it was lost in the retelling. When two other researchers compared false and true stories that had reached the same number of people, the shapes of their spread were almost the same. False stories were catching more often, like a more infectious cold, but they didn’t travel in some special way.

And most people see far less false news than the alarm suggests. One study of what Americans actually read and watched found that news was about 14% of it, and false information about 0.15%. A small group of very active people sees and shares most of it. Of those who do see it, many share it to laugh at it, argue with it, or show which side they’re on, and sharing isn’t believing. The researchers who gathered this evidence put it bluntly: people are more often uninformed than misinformed. The bigger problem may be not the lies people swallow but the true things they never come across.

So which side is right? Mostly, both. Among claims disputed enough to be checked, the false ones really did travel farther and faster, and the reasons, novelty and strong feeling, fit everything else in this chapter. But the false part is a small corner of what most people see, crowded around a few heavy sharers, and an alarm that says “everything online is lies” is itself false, and does its own harm. A person who stops believing anything at all hasn’t become harder to fool. They’ve only stopped being able to learn.

The danger lives in particular moments. It is greatest when something frightening is happening, facts are few, and everyone is anxious and reaching for their phone: an earthquake, a flood, a new disease, an election. That was the night of the Norvic clip. Those are the moments when the feed is most stirred, and when one false forward can send people to the wrong place, or away from the right one.

What can one person do? Not much about the program that orders a feed. But each share is a small decision, and the research gives two plain habits that make it a better one. The first is to notice the stir. When a post makes you angry, frightened or amazed, that is exactly when the feed is working hardest, and exactly when to slow down. The second is to ask the one question that so rarely comes to mind at the moment of the tap: is this true? It takes ten seconds, and it is the whole of Pause and check before you forward. If you can’t tell, the news will carry it if it matters.

The same questions apply to the news itself, the kind with reporters and editors behind it, which is far more reliable than a forwarded clip and can still leave you with a false picture of the world in quite a different way.

Found a mistake? Please tell us.