Social proof: meaning and application

A product page with hundreds of reviews often feels more trustworthy than the same page without them. That is because people use the behaviour and judgements of others as a cue for what is sensible. But when does this social information help a choice, when does it backfire, and where does it turn into pressure or deception?

Social proof is the tendency to let behaviour and judgements be influenced by what other people do or think, especially when you are uncertain yourself.

The term became widely known through Robert Cialdini; the phenomenon itself has been studied experimentally since the 1930s. The effect depends on the reference group, the credibility of the signal and whether the behaviour shown is desirable. In the EU, fake reviews and misleading presentation of reviews count as unfair commercial practices.

The term social proof became widely known through the book Influence by social psychologist Robert Cialdini (Cialdini, 1984), which describes six basic patterns of influence. Cialdini gave the phenomenon a well-known name, but he was not the first to study it: social psychologists had been examining since the 1930s how other people's judgements shape perception and choice. An overview of all the principles is available in Influence: Cialdini principles. This article explores one principle in depth: mechanism, conditions and limits.

• Sherif (1935) had people in a dark room estimate how far a stationary point of light, which appears to move through an optical illusion, moved. In groups, estimates converged on a shared reference point: a group norm of its own (Sherif, 1935).
• Asch (1956) placed one participant opposite a unanimous majority of confederates who gave visibly wrong answers when comparing line lengths (Asch, 1956). The task was clear, yet would people go along?
• Milgram, Bickman and Berkowitz (1969) had groups of different sizes look up on a busy street. The larger the group, the greater the proportion of passers-by who adopted the behaviour (Milgram et al., 1969).

A meta-analysis of 133 line-judgement studies from 17 countries found that conformity in US studies declined after the 1950s and was higher in collectivist countries (Bond & Smith, 1996). Conformity is not a fixed constant.

Deutsch and Gerard (1955) varied Asch's experiment, including anonymous answers and face-to-face answers (Deutsch & Gerard, 1955). Their work gave rise to a distinction that is still used:

• Informational influence: you adopt others' behaviour or judgement as a cue about reality. If many people buy something, it may be good.
• Normative influence: you adapt to stay connected or avoid disapproval, even when you have doubts.

Social proof rests mainly on informational influence, especially when you can judge little yourself. In practice the two overlap: a review score informs, but 'everyone already has this' can feel like an expectation.

Goldstein, Cialdini and Griskevicius (2008) compared a standard environmental card in hotel rooms with cards stating that most guests reuse their towels, a descriptive norm. The norm message led to more reuse, and the version referring to guests in the same room was strongest (Goldstein et al., 2008).

Bohner and Schlüter (2014) repeated this in two German hotels (724 and 204 observations). Both kinds of message increased reuse compared with no message, but the norm messages were not more effective than the standard message, and the effect of the reference group was inconsistent (Bohner & Schlüter, 2014). Reuse was already much higher there: roughly 70 to 90 per cent, against 35 to 50 per cent in the US studies.

The lesson: a norm message can work, but an effect in one context guarantees nothing for another.

Online, social proof appears as reviews, star ratings and notices such as 'frequently bought'. A meta-analysis of 51 studies found an average sales elasticity of 0.236 for the number of reviews and 0.417 for the average rating: 1 per cent more reviews went together with 0.236 per cent more sales on average. The associations were stronger for products that are hard to try beforehand and for reviews on independent platforms (You et al., 2015). These are largely associations, not proven causes.

Numbers have their own dynamics. In experiments by Powell et al. (2017), participants under certain conditions chose the product with more reviews, even when a statistical model judged the product with fewer reviews likely to be better (Powell et al., 2017). See also What is a compelling customer voice in a target group profile?

The role of similarity is a working hypothesis; test it for your target group.

Cialdini (2003) pointed out that campaigns stressing how often undesirable behaviour occurs implicitly say: lots of people do this (Cialdini, 2003). Norm messages work better when what people do and what they approve of align.

In a field experiment, Schultz et al. (2007) found that comparing households with average neighbourhood energy use made heavy users save, but made frugal households use more: a boomerang effect. Adding a message of approval or disapproval removed it (Schultz et al., 2007).

Online, similar logic seems plausible: a counter showing 'two buyers' for a niche model also communicates a norm, but an unfavourable one. That is an editorial inference, not a separately researched effect.

Social proof informs; pressure and deception steer around a person's own judgement. Signs include invented numbers, manipulated reviews and notices that create time pressure.

Fake reviews are not a fringe issue. Luca and Zervas (2016) found that about 16 per cent of restaurant reviews on Yelp were filtered as suspicious; restaurants with a weak reputation were more often involved in fraud (Luca & Zervas, 2016).

Directive (EU) 2019/2161 amended, among others, the Unfair Commercial Practices Directive. If a trader provides access to consumer reviews, information on whether and how it ensures they come from genuine users or buyers is material. Submitting or commissioning false reviews and misrepresenting reviews are on the list of prohibited practices (Directive (EU) 2019/2161). This is not legal advice: have your application legally reviewed.

This plan is editorial advice, not a validated protocol.

1. Describe the choice and the uncertainty. What does the target group find hard to judge?
2. Define the reference group. Who does the target group compare itself with? Research, do not assume.
3. List genuine signals. Which reviews are verified and which numbers are traceable?
4. Explore the meaning. Does 'frequently bought' suggest reliability or mass product? See How do you measure associations?
5. Pretest. Fix outcome measures in advance, including trust and returns. See From target group insights to testable messages
6. Check the conditions. Have legal and ethical points reviewed.

This example is fictional and has no results.

Situation. An online shop for refurbished smartphones and laptops wants to show review scores and a 'frequently bought this month' notice.

Decision question. Does this help hesitant buyers, and which form do they find credible?

Available information. In-house reviews, not all linked to a purchase; sales figures per model; customer questions about warranty and battery condition.

Approach. The team first makes visible how reviews are checked and labels only reviews from verified buyers. Interviews and an association task explore which reference group is recognisable and whether 'frequently bought' suggests reliability or 'cheap and gone fast'. A pre-specified test follows: no social information, verified reviews, reviews with numbers. Outcome measures: purchases, returns and questions about trust.

Possible interpretation. Numbers that help for popular models need not help niche models with a low count. Timers and 'currently viewed by' notices are outside the test.

Next step. Compare results per product group; publish the review policy.

• Treating one study as a law: the hotel study did not simply replicate.
• Using averages as predictions: an elasticity from a meta-analysis does not predict what happens to your product.
• Assuming the reference group instead of researching it.
• Communicating a negative norm: 'many people forget this' normalises forgetting.
• Showing low or unchecked numbers.
• Skipping the legal check. See also What does evidence mean in a target group profile?

People use what others do as information, especially under uncertainty. That makes social signals useful but not predictable: effects differ by context, negative norms can backfire and fake reviews are prohibited in the EU. So first research which reference group and signals your target group finds credible, and then test.