Why We Follow the Crowd: Social Proof, Herding and the Wisdom of Crowds
Imagine arriving in an unfamiliar city and choosing between two neighbouring restaurants. One is nearly empty, while the other has a queue stretching onto the pavement. You know nothing about either menu, but the queue immediately becomes information: perhaps local customers know which restaurant is better, cheaper or more reliable. Choosing the crowded restaurant would not necessarily be irrational conformity. It could be a sensible attempt to learn from people who may know something you do not.
This simple example captures much of why people follow the crowd. Human beings constantly use other people’s behaviour as evidence. We watch which train platform fills when an announcement is unclear, which software experienced colleagues adopt, which doctor several trusted people recommend and how others react when an unfamiliar situation suddenly appears dangerous. Social information saves time because individuals cannot personally investigate every choice from the beginning.
The problem is that visible agreement does not always reveal how much genuine information lies behind it. Ten people may have reached the same conclusion independently after seeing different evidence, or nine of them may simply be copying the first person. Both situations produce a crowd of ten, yet the informational value is radically different. Modern research on social learning and information cascades therefore treats crowd-following as neither inherently intelligent nor inherently foolish; its reliability depends on how the crowd formed.
This distinction also separates crowd-following from a simpler picture of conformity. People sometimes align with groups because they want belonging or fear rejection, but they also follow others because other people may genuinely possess useful knowledge. A 2024 systematic review of contemporary conformity research found robust social influence across many settings while emphasising that situational context strongly affects its strength and form. Digital technology has also created new environments in which popularity, agreement and social reactions are continuously visible.
The useful question is therefore not simply, “Am I following the crowd?” It is what does the crowd actually know, and how independently did its members come to know it?
Social proof becomes powerful when people are uncertain
Other people’s behaviour matters most when our own information is weak. If a question has an obvious answer and we are highly confident, a crowd usually provides less additional value. When the environment is unfamiliar or ambiguous, however, the actions of others can become a substitute for information we do not possess.
A tourist entering an unfamiliar railway station may follow commuters who appear to know where they are going. A junior employee may observe how experienced colleagues behave during a formal meeting because nobody has explicitly explained the organisation’s unwritten rules. A shopper choosing an unfamiliar product may use ratings because personally testing every alternative would be expensive and time-consuming. In each case, observing others reduces uncertainty.
Psychologists often describe this broad process as informational social influence, while economists study related processes under social learning. The underlying logic is straightforward: another person’s action may reveal private information. If someone chooses restaurant A over restaurant B, an observer may reasonably infer that the first person saw, heard or experienced something favourable about A.
Similarity can make that information appear more relevant. A first-time parent may give particular weight to recommendations from other parents with children of a similar age, while an experienced cyclist may care more about bicycle reviews from other serious riders than from occasional users. A tourist may follow local commuters rather than another visibly confused visitor because local people appear more likely to possess relevant knowledge.
This does not mean similar people are necessarily better informed. Similarity is a heuristic for relevance, not proof of expertise. A group of first-time investors can watch one another and create the appearance of collective confidence even when none has strong information. The same thing can occur inside professional communities when people rely on shared assumptions rather than independent evidence.
Uncertainty also makes visible popularity attractive because popularity compresses complicated information into a simple signal. Ratings, queues, bestseller labels, download counts, follower numbers and “trending” indicators allow people to reduce hundreds of possible choices to a manageable shortlist. In a world of information abundance, that filtering function is genuinely useful.
Yet popularity measures behaviour, not necessarily quality. A restaurant may be crowded because it appeared in a viral video. A book may become a bestseller because of a highly effective marketing campaign. An app may dominate because early adoption created network effects that made switching costly. A social-media post may attract enormous engagement because people are angry with it rather than persuaded by it.
This creates the possibility of self-reinforcing popularity. People choose an option because it already looks popular; their choice increases its popularity; later observers then encounter an even stronger popularity signal. The crowd begins influencing the evidence that future members of the crowd will use.
At that point, social learning can turn into herding.
Information cascades explain how a large majority can rest on surprisingly little evidence
An information cascade can occur when people observe earlier choices and decide that those choices contain enough information to outweigh their own private signal. Once this happens, later individuals may rationally imitate the visible majority even if their own evidence points in another direction. The result can be a long sequence of identical behaviour generated from only a small amount of original information.
Suppose the first customer choosing between two unfamiliar products receives a weak private signal favouring product A and buys it. A second customer also receives a weak signal favouring A and makes the same choice. The third customer privately receives evidence favouring B, but sees two earlier people choose A and concludes that their combined information probably outweighs their own. The third person buys A, and later customers now see three apparently consistent choices even though the third added no new evidence supporting A.
A large crowd can then form around information that originated with only the first few decisions. The later choices increase the visible majority without proportionally increasing the independent evidence behind it. This is why the size of a crowd and the amount of information contained in that crowd are not equivalent.
Bikhchandani, Hirshleifer, Tamuz and Welch’s modern review of information cascades describes social learning as capable of producing efficient information aggregation but also inaccurate mass behaviour, fragility and—in strong cascades—a situation where later private information stops influencing observable behaviour. Their review reflects more than three decades of research on how individuals learn from one another in markets and social settings.
Cascades can also be fragile. If the crowd formed from relatively little original evidence, one sufficiently strong piece of public information can reverse behaviour quickly. What looked like overwhelming social certainty may disappear because much of the majority was imitation rather than independent conviction.
This is one reason markets can display herding. Investors observe price movements and other traders’ decisions because those behaviours may reveal information. Buying after others buy can therefore be rational when earlier traders genuinely know more. Yet if each trader increasingly treats rising prices themselves as evidence, buying can generate more buying and create momentum that becomes partly detached from underlying fundamentals.
The same structure appears outside finance. A neighbourhood becomes fashionable partly because people see other people wanting to live there. A collectible becomes valuable partly because others expect future buyers to value it. A technology standard can dominate because widespread adoption makes compatibility more useful, even if competing technologies were initially similar. Herding therefore does not always imply a mistake; network effects can make popularity itself economically important.
Consumer ratings provide another example. An early cluster of positive reviews can encourage more purchases, which generate more opportunities for positive reviews and push the product higher in search results. The product’s later dominance may reflect genuine quality, early luck, visibility advantages or some combination of all three.
The central analytical question is how many judgments are genuinely independent. Ten independent customers who separately inspected a product and reached the same conclusion provide much stronger evidence than ten customers who bought it because it already had excellent ratings. Social behaviour becomes increasingly informative when each person adds new information and increasingly circular when each person mainly reacts to the visible behaviour of previous people.
This distinction matters enormously in journalism and online information. Seeing the same claim on twenty websites can create the impression of twenty confirmations, but all twenty articles may ultimately trace back to one unnamed source, press release or social-media post. Counting repetitions without tracing source lineage turns an information cascade into apparent corroboration.
Independent verification is therefore more valuable than numerical repetition.
People also follow crowds because belonging and reputation matter
Not all crowd-following is about information. Human beings belong to families, friendship groups, workplaces, professions, religious communities, nations, political groups and countless smaller social networks. Clothing, speech, rituals, tastes and behaviour can communicate membership, making conformity partly an expression of identity rather than a judgment about objective truth.
This is not necessarily superficial. Shared conventions make cooperation easier and allow people to communicate belonging quickly. Wearing appropriate clothing at a funeral, following professional etiquette or participating in a community ritual can show respect without requiring an individual to independently reinvent the correct behaviour.
The problem becomes more serious when standing apart carries enough social cost to suppress important private information. Someone may believe a workplace decision is flawed but remain silent because every senior colleague appears enthusiastic. A teenager may participate in risky behaviour because refusal threatens belonging. A political supporter may avoid questioning a claim associated with their own side because dissent could be interpreted as disloyalty.
A particularly important phenomenon is pluralistic ignorance. This occurs when many people privately reject or question a norm but mistakenly believe that most others genuinely support it. Because everyone observes public behaviour rather than private thoughts, people may continue performing a norm that surprisingly few members actually endorse.
Imagine a meeting in which several employees privately believe a proposal is unrealistic. The manager expresses enthusiasm, nobody immediately objects and each doubtful employee interprets everyone else’s silence as agreement. They remain silent too, strengthening the apparent consensus. By the end of the meeting, the group looks highly confident despite containing substantial private doubt.
This explains why anonymous polls, private voting and independent estimates can sometimes reveal far more disagreement than public discussion suggests. Visible behaviour is evidence about what people are willing to express under current social conditions, not perfect evidence about what they privately believe.
The distinction between public behaviour and private acceptance is well established in conformity research. People can outwardly align with a group while retaining their original view, or social information can genuinely change the belief itself. That difference matters because a unanimous public group can be much less psychologically unanimous than it appears.
Fear of embarrassment adds another layer. When everyone else moves confidently in one direction, a dissenting person can begin asking whether the problem lies with their own understanding. Even when they remain privately unconvinced, being the only visible dissenter creates reputational risk. Silence then becomes easier than testing whether the majority actually possesses better evidence.
Leadership can amplify these effects because people do not observe every member of a crowd equally. Recognised experts, senior managers, celebrities, influencers and highly connected individuals receive disproportionate attention. An early statement from one high-status person can shape what everyone else later interprets as the group’s natural consensus.
This is why good leadership does more than state conclusions confidently. When leaders explain the evidence, acknowledge uncertainty and allow others to form independent judgments before announcing their own preference, they preserve information that might otherwise disappear into conformity. A leader who asks everyone to copy their confidence may produce rapid coordination, but not necessarily an intelligent group.
Crowds become wise when they aggregate different information rather than copy themselves
The idea that crowds can be wrong should not lead to the opposite myth that individuals are generally wiser alone. Under the right conditions, aggregating multiple judgments can produce remarkably accurate results. The phenomenon popularly called the wisdom of crowds occurs because different people possess different information and make different errors, allowing some of those errors to cancel when their judgments are combined.
Independence and diversity are therefore crucial. A review in Nature Reviews Psychology emphasises individual heterogeneity as a central ingredient in collective intelligence and discusses mechanisms that allow groups to combine specialised knowledge or differently informed judgments. Collective performance improves not merely because a group is large, but because members contribute sufficiently different information and that information is aggregated effectively.
A useful crowd therefore resembles many partially independent sensors rather than a hall of mirrors. If one person systematically underestimates while another overestimates, averaging their estimates may reduce error. If everybody receives the same misleading information and applies the same reasoning, increasing the number of people can simply reproduce the same bias more confidently.
Experimental research illustrates this principle. Independent estimates can combine into highly accurate collective judgments, whereas social influence can reduce diversity and thereby weaken the crowd’s informational advantage. Studies have also shown that structured interaction can sometimes improve collective intelligence when people contribute different knowledge rather than simply converging prematurely.
This creates an apparent paradox. People use social information because learning from others can improve individual judgment, yet excessive social learning makes the group’s opinions more correlated. As more people begin copying one another, the independence that made the crowd useful begins disappearing. Research modelling cultural diversity and collective intelligence describes precisely this tension: social transmission can reduce within-group diversity and undermine the wisdom that aggregation originally produced.
The solution is not complete isolation. Groups often benefit from exchanging information, identifying expertise and coordinating specialised work. The challenge is preserving enough independent thinking before convergence occurs.
This is why prediction systems and good decision procedures often collect judgments separately before showing people the aggregate. A team might ask each member for a forecast before discussion, then compare estimates and examine why they differ. The group receives the benefit of social learning after capturing private information rather than allowing the first speaker to reshape everyone else’s answer.
Expertise can also be weighted rather than treating every voice identically. If one participant possesses specialised knowledge directly relevant to the problem, giving that information greater weight can improve collective performance. But expertise itself should be established by track record, relevant knowledge or evidence rather than by popularity.
The wisdom of crowds therefore depends on more than head count. A thousand people who independently possess weak but partly informative signals can sometimes outperform one highly confident individual. A million people repeating the same unsupported claim can remain wrong.
The most important resource in a good crowd is not size alone. It is informational diversity.
Digital platforms make popularity visible before people have formed their own judgment
Digital platforms dramatically increase the amount of social information surrounding almost every decision. Before reading a video, article or post, users may already see its view count, number of shares, rating, comments and whether an algorithm has labelled it “trending.” The response of the crowd becomes part of the object being judged.
These signals can be useful. A product with thousands of consistently strong reviews may deserve attention, and a technical answer endorsed by many experienced users can help someone identify a likely solution quickly. The problem arises when users mistake engagement for truth, quality or consensus.
Engagement measures attention. People click, share and comment for many reasons, including disagreement, outrage, humour, identity signalling and curiosity. A highly engaged post may therefore be influential without being widely believed.
Research on political social-media content illustrates how strongly engagement can favour emotionally divisive material. Rathje, Van Bavel and van der Linden analysed more than 2.7 million posts from U.S. news organisations and members of Congress and found that posts referring to political out-groups were substantially more likely to be shared or retweeted than posts focusing on political in-groups. The study demonstrates that what becomes visible online can partly reflect what generates engagement rather than what users consider most accurate or socially representative.
Online popularity is also vulnerable to manipulation. Bots, coordinated campaigns, purchased followers and organised review activity can manufacture the appearance that many independent people spontaneously support something. Algorithmic amplification introduces another complication because users do not see a neutral sample of everybody’s behaviour; they see behaviour selected by a ranking system designed according to particular objectives.
A “trending” label therefore tells users that something is receiving attention according to the platform’s measurement system. It does not automatically reveal whether the attention is positive, whether the underlying claim is correct or whether the visible audience represents the wider population.
The same caution applies to star ratings. A numerical average hides distribution, timing, selection and context. A product with 4.8 stars from enthusiastic early adopters may suit a specialist audience while being unsuitable for most buyers. Recent detailed reviews discussing the exact use case may contain more useful information than the aggregate score.
Digital environments also accelerate cascades because people can observe thousands of previous choices almost instantly. One early viral post can become the source for hundreds of derivative posts, reactions and articles, creating the impression of a broad information base when much of the activity ultimately traces back to one source.
The critical skill is therefore source independence. Before treating online consensus as powerful evidence, ask whether different voices have independently verified the underlying claim or merely reproduced something already popular.
Following the crowd in emergencies can be adaptive, not evidence of “mass panic”
Crowd-following becomes especially interesting in emergencies because information is scarce and decisions can be urgent. If several people suddenly move toward an exit after hearing something you did not hear, following them can be sensible. Waiting to independently verify a fire before evacuating may be much more dangerous than using other people’s behaviour as an early warning signal.
Popular accounts often describe emergency crowds as irrational, contagious and prone to panic. Modern crowd research presents a considerably more nuanced picture. Reviews of emergency behaviour find that indiscriminate mass panic is relatively uncommon and that cooperation, mutual support and shared identity frequently emerge during disasters and other collective emergencies.
This does not mean crowds always respond correctly. People can follow an incorrect evacuation route, spread a false alarm or copy behaviour based on incomplete information. A few highly visible actions can become powerful social signals precisely because individuals have little time to investigate what caused them.
Reliable communication is therefore crucial. Emergency authorities who provide clear, credible and timely information reduce the need for people to infer danger entirely from the movements of strangers. Research on crowd psychology and emergency management also emphasises communication and shared identity as important factors supporting coordination and cooperation.
The correct lesson is not “never follow a crowd during an emergency.” Sometimes the crowd possesses information you do not, and sometimes immediate collective movement is exactly what safety requires. The better principle is that public communication should give crowds access to reliable common information so that copying behaviour is less dependent on rumour.
This applies to breaking news more broadly. During rapidly developing events, people often encounter reactions before verified explanations. Shares, screenshots and eyewitness claims can spread faster than professional verification, creating strong social proof around information whose origin remains uncertain.
When stakes are high, the usefulness of the crowd depends heavily on whether its apparent consensus arose after independent verification or before it.
Good judgment means learning from the crowd without surrendering to it
Advice about herd behaviour sometimes replaces one simplistic rule with another: instead of “follow the crowd,” it says “always think independently.” That is not much better. A person who automatically rejects every majority is still allowing the majority to determine their behaviour; the response is simply reversed.
True independence does not require ignoring social evidence. It means treating other people’s behaviour as one source of information and asking how much weight that information deserves. When experienced people have independently examined a problem and largely agree, updating toward their judgment can be highly rational. When a viral crowd is copying one influencer who provided no evidence, numerical popularity should carry much less weight.
Four questions are particularly useful. Do the people making the choice possess knowledge relevant to the decision? Did they reach their conclusions independently? Could they all be responding to the same misleading source or social pressure? And how serious would the consequences be if the crowd turned out to be wrong?
The last question changes how much verification is sensible. Choosing a busy restaurant involves limited downside, so copying local customers may be efficient. Investing a large portion of savings because a stock is trending online carries much greater cost, making independent investigation more important.
It is also useful to identify the first non-social reason supporting the crowd’s behaviour. If a product is popular, is there credible performance evidence explaining why? If everyone is leaving a building, is there an alarm, smoke or authoritative instruction? If a claim is spreading rapidly, can it be traced to a reliable primary source?
When no reason can be found beyond “many people are doing it,” the evidence is weaker than the size of the crowd makes it appear.
Consumers can apply the same principle by looking beneath aggregate ratings. Detailed reviews from users with similar needs, independent tests, common negative complaints and long-term experience often reveal more than a star average. Investors can distinguish price momentum from new information about underlying assets. Professionals can form preliminary judgments before seeing colleagues’ answers.
Groups can also protect collective intelligence through design. Gather individual estimates before discussion, invite people with different expertise, trace claims to independent sources and avoid letting status determine who speaks first. These practices preserve the diversity that makes group judgment valuable.
The objective is not to prevent people from influencing one another. Social learning is one of humanity’s most powerful cognitive tools. Civilisation itself depends on the ability to acquire knowledge from people we have never personally observed proving every claim.
The danger begins when social information becomes circular. One person acts, a second copies, a third interprets the first two as independent evidence, and eventually thousands of people treat the size of the crowd as proof that the crowd must have had a good reason. The appearance of knowledge expands while the amount of independent knowledge may barely change.
That is why the crowd can be both wise and foolish without contradiction. A diverse group pooling independent information can outperform its individual members. A large group copying the same initial signal can amplify an error much faster than one person could.
People follow crowds because doing so often works. Other humans really do possess information, experience and local knowledge we lack, and social coordination makes life possible. The mistake is not using social proof; it is failing to ask what produced the proof.
The best judgment therefore lies between isolation and imitation. Learn from other people, but distinguish independent evidence from repetition. Treat expertise differently from popularity, and treat popularity differently from truth. When uncertainty is high and the crowd knows more, following can be rational. When the crowd is mainly watching itself, the strongest-looking consensus may contain surprisingly little information.
A crowd becomes most useful when its members bring different knowledge to the decision. It becomes most dangerous when everyone assumes somebody else already knows why they are moving in the same direction.
