Everyone agrees: social proof and manufactured consensus

Series: How minds are moved · Foundations: how persuasion works · 5 of 7

In 1951 the psychologist Solomon Asch showed students a line and asked which of three other lines matched it. The answer was obvious. But the other people in the room, who were secretly working with Asch, gave the same wrong answer out loud, one after another. About 75% of the real participants went along with the group at least once. Seventy years later, the “people in the room” can be thousands of accounts run by a handful of operators, and governments of every kind have been caught building them.

Contents · 8 sections

TL;DR #

• We judge what is right partly by what others seem to think. Psychologists call this social proof, and it is strongest when we are unsure.
• Asch’s experiments showed how far people will go along with a confident group, even against their own eyes.
• Fake consensus is now an industry: astroturf campaigns, paid reviews, bot networks and sockpuppet accounts.
• The record includes documented operations attributed to Russia, China and the United States, a reported Israeli-government-funded campaign, and commercial firms.
• Most of these operations reached small audiences. Their real power is the illusion that “everyone” agrees.

Why the crowd feels like evidence #

Robert Cialdini, in Influence, calls it the principle of social proof: we decide what is correct by looking at what other people do. Most of the time this is a sensible shortcut. If everyone is leaving a building, leave. If a restaurant is full, it is probably good.

Cialdini adds the condition that matters: “Ah, uncertainty—the right-hand man of the principle of social proof.” When we are unsure, in a new place, on a new topic, in a crisis, we lean on the crowd most. And we lean hardest on people who seem similar to us.

In Asch’s line experiments, each participant faced 12 critical trials where the group gave a wrong answer. About three quarters conformed at least once, and roughly a third of all answers on those trials followed the wrong majority. A quarter never gave in. And when even one other person in the room gave the right answer, conformity dropped sharply. That last detail is the most hopeful one in the whole literature: one dissenting voice breaks the spell.

Manufactured consensus #

If the crowd is evidence, then faking the crowd is a way to fake evidence. The US senator Lloyd Bentsen is credited with the word “astroturf” in 1985, for letter-writing campaigns that pretended to be grassroots. The methods have changed; the idea has not.

Fake reviews #

Product reviews are social proof with stars. They have been bought and sold for years, and regulators have caught up only recently. In August 2024 the US Federal Trade Commission adopted a rule banning fake reviews and testimonials, including ones written by AI, the buying of positive or negative reviews, and the selling of fake followers or views. It allows civil penalties for each violation. The UK’s 2024 consumer law added similar bans. These rules exist because the practice was widespread.

State sockpuppets, East #

• Russia: the US Special Counsel’s report of 2019 described the Internet Research Agency in St Petersburg, which ran fake American personas on social media. Twitter identified 3,814 accounts it linked to the agency, and Facebook estimated that its content may have reached up to 126 million people. How much it changed anyone’s vote is a separate question, and research since has found the effects were small.
• China: in 2017 researchers Gary King, Jennifer Pan and Margaret Roberts estimated that the Chinese government fabricates about 448 million social media posts a year, written by government employees posing as ordinary citizens. Their finding was that the posts mostly avoid arguing and instead distract, cheerlead and change the subject. In August 2023 Meta removed 7,704 Facebook accounts and 954 pages from a network called Spamouflage, which it linked to individuals associated with Chinese law enforcement and called its largest single takedown. Meta also said the network struggled to reach real people beyond its own fake echo chamber.

State sockpuppets, West #

• In 2011 the press reported that US Central Command had contracted “persona management” software, Operation Earnest Voice, to let operators run multiple online identities in foreign languages. The stated rule was that it would not target Americans.
• In August 2022 the Stanford Internet Observatory and the analytics firm Graphika published Unheard Voice, an analysis of accounts that Twitter and Meta had removed. The Twitter data alone covered 146 accounts and nearly 300,000 tweets from 2012 to 2022, promoting pro-Western narratives in the Middle East and Central Asia, with fake personas and fake news sites. The authors called it the most extensive case of covert pro-Western influence operations on social media reviewed by open-source researchers. Most posts got almost no engagement. In November 2022 Meta said it had found links to individuals associated with the US military. In September 2022, after the takedowns, the Pentagon’s policy chief Colin Kahl ordered a review of the military’s clandestine online influence operations, The Washington Post reported.
• In June 2024 Reuters reported that the US military had run roughly 300 fake accounts posing as Filipinos, from spring 2020 to mid-2021, to cast doubt on China’s Sinovac vaccine. A Pentagon spokesperson acknowledged that it had spread content about Sinovac’s “safety and efficacy”. The Pentagon later said an initial review found the military “not responsible for the troubling social media content”, while a non-public document given to Manila, seen by Reuters, admitted “missteps” in its COVID messaging.
• In June 2024 the New York Times reported that Israel’s Ministry of Diaspora Affairs had paid about 2 million dollars to a Tel Aviv marketing firm for a campaign of fake accounts aimed at US lawmakers over the Gaza war. The minister, Amichai Chikli, denied it. Meta and OpenAI had already attributed the activity they removed to that firm, STOIC.

At home: Romania, 2024 #

On 6 December 2024 Romania’s Constitutional Court annulled the whole presidential election, after its first round and before the runoff, following declassified intelligence notes that described a sudden wave of coordinated TikTok promotion for the candidate Călin Georgescu. Russia denied any role, and who operated the network has not been publicly established. Later in December, Romanian media, citing the tax authority, reported that one campaign used in that wave had been paid for by a mainstream Romanian party through a PR agency, with influencers who were supposed to promote “a candidate” in general, and that the hashtag was then taken over. The agency said its campaign had been hijacked. On 10 November, my news engine saw a press conference about the long-delayed official report on the annulment run under opposite headlines in Gândul and Capital. The official report is still unpublished, and the full story is contested.

What it feels like from the inside #

Social proof feels like relief. You are not alone, you are not crazy, you are on the right side. Seeing a thousand likes under a post makes it feel safe to agree, and seeing a thousand angry replies makes it feel dangerous to disagree. Disagreeing with a visible majority produces real discomfort; Asch’s participants described doubting their own eyesight.

The trick of manufactured consensus is that it works on both feelings at once: it makes one view look popular and the other look lonely.

What it can and cannot do #

• Most documented bot and sockpuppet networks had small real audiences. Spamouflage and Unheard Voice both largely talked to themselves.
• Their effect is probably larger on journalists, politicians and moderators who watch online trends and treat them as public opinion, than on ordinary readers.
• Fake reviews do change purchases; that is why firms pay for them and regulators fine them.
• Real consensus exists too. Many people agreeing is not proof of manipulation. The question is whether the many are real and independent.

How to notice it #

1. Are these accounts real? Look for new accounts, no history, stock photos, identical wording.
2. Count independent voices, not posts or likes.
3. Did the “wave” appear all at once, on one topic, from accounts that were quiet before?
4. Are reviews all five stars or all one star, short, vague and posted in bursts?
5. Am I agreeing because I checked, or because everyone seems to agree?
6. Find one person you trust who disagrees, and listen to them. One dissenter breaks the Asch effect.

Frequently asked questions #

What is social proof? #

It is the tendency to judge what is correct by what other people seem to think or do. Robert Cialdini named it in Influence. It is strongest under uncertainty and when the others seem similar to us.

What is astroturfing? #

It is a campaign made to look like spontaneous public support when it is organised and paid for by a company, party, government or interest group. The name is a play on a brand of artificial grass.

Do Western governments run fake accounts too? #

Yes, this is documented. Examples include the US Operation Earnest Voice contract, the pro-Western network analysed by Stanford and Graphika in 2022, and the US military’s 2020 to 2021 anti-Sinovac campaign in the Philippines reported by Reuters.

Do bot networks actually change opinions? #

The evidence suggests direct effects on ordinary users are usually small, because most fake accounts reach few real people. Their bigger effect may be on what journalists and officials believe the public thinks.

The lesson #

The crowd is often right, and following it is not foolish. But a crowd can now be built, cheaply, by states of every bloc, by companies and by campaigns. Before you treat agreement as evidence, check that the people agreeing are real and independent. And if you are the one person in the room who sees the line differently, say so: Asch showed that one voice changes the whole room. Next in the series: how a war needs an enemy, and the ten principles that build one.