Hand 599 Americans and Britons ten real conspiracy theories and ask two separate questions: do you personally believe this, and what percentage of other people do you think believe it? Then take a second batch of a thousand more participants, have half of them read an actual blog post arguing for one of those theories before answering, and watch what happens to both numbers at once. That’s the design behind Everyone Knows It’s a Hoax: The Role of Perceived Descriptive Norms in Conspiracy Beliefs, published online Sept. 9, 2026 in the European Journal of Social Psychology by Jan-Willem van Prooijen of Vrije Universiteit Amsterdam with co-authors Paul K. Bergmann, Vladimíra Čavojová, and Jakub Šrol, and covered Oct. 23, 2026 by PsyPost. Reading the blog post didn’t just move personal belief from an average of 49.46 to 61.78 out of 100. It moved the estimate of what everyone else believed almost exactly as far, from 51.00 to 62.64. The same nudge that convinced someone also convinced them, instantly, that they’d joined a crowd that already agreed. It’s close to a live demonstration of something a different research team found the year before: Pennycook, Binnendyk, and Rand found that people who believe a given conspiracy theory think they’re in the majority 93% of the time — when, on average, only 12% of people actually share the belief. Gustave Le Bon named the mechanism behind both numbers in The Crowd back in 1895: “the unanimity of numerous witnesses is one of the worst proofs that can be invoked to establish a fact.”

The Quick Verdict

Question	What we found
What was studied?	Whether believing “everyone agrees with me” drives conspiracy belief, and whether correcting that perception changes it
Who ran it?	Jan-Willem van Prooijen (Vrije Universiteit Amsterdam), with Paul K. Bergmann, Vladimíra Čavojová, and Jakub Šrol
Where and when?	European Journal of Social Psychology, published online Sept. 9, 2026; covered Oct. 23, 2026 by PsyPost
Sample	Six preregistered studies, 2,906 US and UK adults
Core finding	Reading conspiratorial content raised personal belief and perceived consensus by nearly matching amounts, in the same people, at the same time
The intervention that worked	Showing participants real statistics on how few people actually endorse a claim cut both their perceived-consensus estimate and their own belief
The corroborating study	Pennycook, Binnendyk, and Rand, Personality and Social Psychology Bulletin 52(8): believers think they’re in the majority 93% of the time; actual majority support averages 12%
What Le Bon already argued	The Crowd (1895): a suggested, contagious impression gets mistaken for independently observed agreement — “the unanimity of numerous witnesses is one of the worst proofs”

The Quick Version

Two research teams, working independently a year apart, converged on the same strange gap: people who believe a conspiracy theory think almost everyone agrees with them, when almost nobody does. One team found the false-consensus number — 93% perceived agreement against 12% actual agreement. The other found the mechanism that produces it — reading a single piece of persuasive content shifts your own belief and your estimate of the crowd’s belief together, as one motion, not two. Gustave Le Bon described this in 1895 as a feature of physical mobs. It turns out it also happens inside one person’s head, alone, reading a blog post.

What the Studies Actually Found

Van Prooijen’s team started with a baseline question nobody had pinned down cleanly: does perceived consensus track actual belief, or does it run ahead of it? Across the first two studies (599 participants, US and UK), actual average belief in ten real conspiracy theories came out to 33.09 and 34.16 out of 100. Estimates of what the average other person believed came out higher — 39.85 and 42.30. Among people who scored as strong believers specifically, the correlation between their own belief and their estimate of others’ belief was .65 in the US sample and .71 in the UK sample. The stronger someone’s own conviction, the more they assumed it was shared.

Then came the experiment that explains why. Half of roughly a thousand new participants read a real blog post arguing for a specific conspiracy theory before answering the same two questions; the other half didn’t. In the US sample, reading the post pushed personal belief from 49.46 to 61.78, and pushed the estimate of others’ belief from 51.00 to 62.64 — almost the identical jump, in the same people, from the same ten minutes of reading. The UK sample showed the same pattern: 41.99 up to 59.78. A third study tried the cheap version of the same idea — just telling people vaguely that “many” versus “few” others agreed — and found no significant effect. Vague suggestion didn’t move anyone. Concrete, specific content did.

The intervention study is the one worth sitting with longest. Eight hundred participants were split between seeing an actual bar chart of how many people endorse a set of real conspiracy claims (mostly not many) and seeing no such chart. The group that saw the real numbers estimated the average belief at 20.95 out of 100, versus 38.34 in the no-data group — and their own personal endorsement of the claims came in at 24.06, versus 33.70 among people given nothing to correct the record. Correcting the perception of consensus didn’t just change what people thought about other people. It changed what they believed, period.

That result lines up with a finding Gordon Pennycook, Jabin Binnendyk, and David G. Rand published the year before, across eight studies and 4,181 U.S. adults. First reported online in May 2025 and formally appearing in the August 2026 print issue of Personality and Social Psychology Bulletin — first covered by PsyPost in June 2025 and later by Neuroscience News — their research measured something adjacent: dispositional overconfidence, using tasks on visual perception, numeracy, and cognitive reflection that have nothing to do with politics or conspiracy at all. People who scored as chronically overconfident on those unrelated tasks were the same people most likely to believe conspiracy theories, and most likely to badly misjudge how popular those beliefs were. The specific number: conspiratorial claims were actually endorsed by a majority of respondents only 12% of the time, but believers thought they held majority support 93% of the time. Two teams, two methods — one measuring a general personality trait, one running a live persuasion experiment — landing on the same shape of error from opposite directions.

What Did Le Bon Mean By This?

In The Crowd, Book I, Chapter II, Gustave Le Bon writes: “Treatises on logic include the unanimity of numerous witnesses in the category of the strongest proofs that can be invoked in support of the exactness of a fact. Yet what we know of the psychology of crowds shows that treatises on logic need on this point to be rewritten. The events with regard to which there exists the most doubt are certainly those which have been observed by the greatest number of persons.” He adds: “A crowd scarcely distinguishes between the subjective and the objective. It accepts as real the images evoked in its mind.”

Le Bon’s illustration is a famous naval anecdote: the crew of the frigate Belle Poule, scanning the sea for a lost vessel, convinced themselves they saw it, signal flags and distress boats and all — and dozens of trained sailors later swore to details of a ship that was never there. His explanation wasn’t that the sailors were lying or stupid. It was that “the first perversion of the truth effected by one of the individuals of the gathering is the starting-point of the contagious suggestion.” One person’s mistaken impression doesn’t stay one person’s. It spreads by suggestion, catches by contagion, and by the time it’s done circulating, everyone remembers it as something they independently observed — not something they caught from the person standing next to them.

That’s the mechanism, not the metaphor, behind van Prooijen’s exposure experiment. A single blog post functioned exactly like Le Bon’s first mistaken sailor: it planted an impression, and the mind receiving it couldn’t cleanly separate “this idea seems plausible” from “this idea is widely held.” Both judgments moved together because a mind in the grip of suggestion isn’t running two calculations. It’s running one. The crowd Le Bon described needed dozens of bodies on a deck together. The effect he was describing, it turns out, needed only one screen and one blog post.

Why Does Reading a Claim Make It Feel Like Consensus?

1. A persuasive argument supplies a felt sense of agreement, not just content. Belief and perceived-consensus estimates moved by nearly identical margins because the two aren’t processed as separate questions — accepting an argument and imagining its audience happen in one motion, exactly what Le Bon meant by suggestion spreading as contagion rather than evidence.
2. Vague social proof doesn’t work, but specific content does. Telling people “many” versus “few” others agree moved nothing. Absorbing a vivid, concrete argument did — and got mistaken for popularity.
3. Dispositional overconfidence supplies the fuel before any specific claim shows up. Pennycook, Binnendyk, and Rand’s tasks on perception, numeracy, and cognitive reflection predicted both conspiracy belief and the 93%-vs-12% consensus error, suggesting some people arrive at a claim already primed to mistake their own certainty for the room’s.
4. Correcting the record takes the same currency that created the error: concrete numbers, not reassurance. The chart intervention worked because it fought a specific, vivid impression with an equally specific, vivid fact — not a generic appeal to “most people disagree.”

Why the Study and the Philosophy Rhyme

The honest version of this comparison isn’t “a nineteenth-century crowd theorist predicted a 2026 online experiment.” It’s narrower, and it holds up better for being narrower: Le Bon’s specific claim was that a crowd cannot reliably separate a suggested impression from an independently confirmed fact, and that unanimity born of contagion looks, from the inside, identical to unanimity born of evidence. Van Prooijen wasn’t testing Le Bon. He was testing whether reading persuasive content changes perceived norms. But the result where belief and perceived-consensus moved together, as a single unit, is a near-literal restatement of the distinction Le Bon drew in 1895 between a fact and a feeling that mimics one.

It connects to the Noelle-Neumann piece on the spiral of silence, which found a loud, overconfident minority online can make moderates fall silent — a related mechanism from the opposite angle: Le Bon’s crowd doesn’t just feel unanimous to the people inside it, it can look unanimous to everyone watching from outside too. It sits close to the Sartre piece on hidden bias, which found that what looks like automatic bias often tracks conscious caution instead — a reminder that confident self-assessment and accurate self-assessment are different variables, exactly the gap Pennycook’s overconfidence tasks measure here. It rhymes with the McLuhan piece on texting versus algorithmic feeds, since a single scrollable blog post did in van Prooijen’s lab what an endless feed does at scale: supply a steady stream of first mistaken sailors. And it stands in useful tension with the Tocqueville piece on institutional commitment: specific, concrete institutional commitments hold up better than abstract democratic sentiment, and specific, concrete numbers correct false consensus better than a vague appeal to “most people know better.”

The Gap Le Bon Didn’t See Coming

Here’s where the comparison is more interesting than a clean match.

Le Bon was describing bodies physically assembled — a mob in a square, a jury in a room, sailors on one deck, all able to see and hear each other producing the contagion in real time. He had no framework for a solitary reader, alone with a screen, absorbing a single blog post with no other person present at all. Van Prooijen’s experiment shows the mechanism doesn’t need the crowd’s physical presence anymore. It needs only the content a crowd would have produced, delivered by one post to one reader, and the reader’s own mind supplies the rest — manufacturing a felt sense of “everyone agrees” with no actual everyone in the room.

He also never tested whether his own mechanism could run in reverse. Le Bon’s crowd psychology reads, at points, like a one-way slide: suggestion catches, contagion spreads, and the resulting conviction is treated as basically permanent once formed. He offers no experimental counter-suggestion. The chart intervention that cut belief from 33.70 to 24.06 is a demonstration his framework didn’t anticipate needing: the same contagious mechanism that inflates a false sense of consensus can be interrupted by a more specific, more concrete fact. Suggestion, in his account, is something crowds suffer. The 2026 data suggests it’s also something an intervention can fight with its own weapon.

What This Study Doesn’t Show

Be precise about the size of this. The framing invites more confidence than the data earns in either direction.

Volunteer Online Panels, Not a Random Public Sample

Both studies recruited through online research panels in the US and UK. People willing to sign up for paid psychology studies aren’t a random cross-section of either country, and panel-recruitment effects on conspiracy belief specifically aren’t ruled out by either paper.

A Short-Term Shift, Measured Immediately

The blog-post exposure and the chart intervention were both measured right after the manipulation. Neither study tracked whether the shift held a day later, or collapsed the moment participants moved on to something else.

Ten Chosen Conspiracy Theories, Not Every Conspiracy Theory

Van Prooijen’s team selected a specific set of claims to test. Highly politically charged theories tied to identity or partisanship may behave differently, where motivated reasoning runs deeper than a single chart can reach.

Overconfidence Correlates With Belief — It Doesn’t Fully Explain It

Pennycook, Binnendyk, and Rand’s tasks predict conspiracy belief and consensus misperception; they don’t claim to be the whole story. Overconfidence is one ingredient alongside social context, institutional distrust, and the specific content a person encounters — the same content van Prooijen’s experiment shows can move belief on its own.

What Actually Follows From This

1. “Do most people believe this” is a question worth actually asking, not assuming. The gap between 93% perceived agreement and 12% actual agreement is large enough that checking — a real poll, a real statistic, not a vibe — is one of the cheapest corrections available before adopting a belief because it feels widely shared.
2. Notice when one piece of content is doing two jobs at once. If reading a single argument changes both what you believe and how popular you assume that belief is, that’s the suggestion-and-contagion mechanism Le Bon described, running on a sample size of one.
3. Ask Le Bon’s actual question about any claim that feels overwhelmingly shared: how many people did I actually hear from, versus assume? The sailors of the Belle Poule were certain dozens of people saw what they saw. The number who actually did was closer to one, misread, and repeated.

Le Bon never ran a controlled experiment, and van Prooijen’s team wasn’t out to vindicate a nineteenth-century theory of mobs. But a 2026 study built to test whether perceived consensus drives conspiracy belief — and whether correcting that perception helps — landed on the exact distinction Le Bon insisted on in 1895: unanimity born of a contagious impression and unanimity born of an actual count are not the same thing, and the mind that feels the first rarely notices it’s missing the second.

---

This post discusses a 2026 study published in the European Journal of Social Psychology (van Prooijen, Bergmann, Čavojová, and Šrol, DOI: 10.1002/ejsp.70111), newly covered by PsyPost in October 2026; a related 2025/2026 study published in Personality and Social Psychology Bulletin (Pennycook, Binnendyk, and Rand, DOI: 10.1177/01461672251338358); and Gustave Le Bon’s The Crowd (1895), Book I, Chapter II. Both studies are drawn from online research panels, and neither measures whether belief or consensus-perception changes persist beyond the immediate testing session.