Social proof: what audience claims support and what they do not

“Over 10,000 satisfied customers”

The line appears on countless home pages. It is meant to show that others have already decided — and that you can hardly go wrong. The principle behind it is social proof, and it counts as one of the more reliable rules in marketing.

Less of it is reliable than the reputation suggests. Research separates two forms that practice runs together, and it knows a case in which exactly the same message produces the opposite result.

Two forms that do not do the same thing

Descriptive norm: what others do. “Most guests reuse their towels.” This form describes behaviour. It works because it resolves an unclear situation: if I do not know what is usual, what others do is the best information available.

Injunctive norm: what one ought to do. “Please reuse your towels.” This form evaluates. It acts more weakly on immediate behaviour but carries information the descriptive form lacks: a direction.

The distinction is not academic. It decides whether a message is dependable or unpredictable.

The boomerang: when the norm pulls downward

Schultz and colleagues (2007) hung 287 households in a Californian town their electricity use beside the neighbourhood average. The result ran in both directions.

The d measures how large an effect is; 0.5 counts as medium. So neither shift is a marginal one — the message did as much damage on one side as it did good on the other.

That is not measurement error but the logic of a descriptive norm: it names a target value, and people approach it from both sides. The increase was no longer detectable once the feedback also carried an evaluative mark — a hand-drawn smiley, happy for frugal households, sad for the heavy consumers. In this study the injunctive norm therefore catches what the descriptive one causes — across these 287 households. In the later rollout at scale that did not hold for every group; see below.

Two details make the finding more awkward still. After four weeks the boomerang among the frugal households held (d = 0.59), while the saving among the heavy consumers was no longer significant without the evaluative mark: the harm outlasted the benefit. And although the two norms combined saved more among heavy consumers than the descriptive norm alone (1.72 against 1.22 kilowatt-hours), the difference was not significant. The evaluative mark prevents the boomerang; it does not demonstrably amplify the message.

The same mechanism had already been shown seventeen years earlier. In the fourth experiment of Cialdini, Reno and Kallgren (1990) the same litter lay either scattered or swept into piles. A swept pile signals disapproval — an injunctive norm. Where a confederate drew attention to it, 45 percent littered with scattered litter but only 18 percent with swept. Without the attention-focusing the difference vanished (33 against 29 percent).

Two demonstrations from the same line of research are not an independent replication, though, and Cialdini is a co-author of both papers. The injunctive norm as protection against the boomerang therefore stays weakly supported. It hangs on this one study, one Californian town and 287 households. That the same combination now goes out to millions (see the next section) changes nothing about that: there the two norms are shipped together, and the injunctive share cannot be separated out.

For a shop this means “most customers choose the middle plan” can pull customers who were on the large plan down to the middle one. Whether that is what you want depends on what you are trying to sell — and nobody checks it before the sentence goes on the page.

What became of it at scale

In 2018 the authors published a retrospective on their own study (Schultz et al., 2018). For practice it is the most useful of the papers evaluated here, because it traces the path from a field experiment with 287 households to mass application.

Since 2008 the company Opower has been sending out electricity bills carrying exactly this feedback — a neighbourhood comparison plus an evaluative mark. By now that is more than 60 million households, and every rollout is built as a randomised trial: at a utility serving 100,000 homes, half receive the reports and half do not.

Two numbers from it matter most for shop copy. First: across more than a hundred rollouts the savings run between 2 and 5 percent. That is the order of magnitude a cleanly built normative message delivers in continuous operation — not a lever that overturns a conversion rate. Second: after the reports stopped, the effect halved (Allcott & Rogers, 2014). It did not vanish, but it lives on repetition.

And the boomerang showed up at scale too. A reanalysis of the data (Costa & Kahn, 2013) found a group whose consumption rose in response to the message: politically conservative households above the norm who did not donate to environmental organisations and did not pay for renewable energy. The evaluative mark does not catch the boomerang for everyone. What matters is whether the recipient identifies with the referent group at all — which is what the next section is about.

One open question the authors hand on themselves. Nobody knows which figure makes a strong descriptive norm: their own study gave a mean, other work gives percentages. Early findings suggest a low but rapidly rising figure can motivate while a high but falling one does not. For “over 10,000 satisfied customers” that means: the form of the number is untested, in this literature too.

What 91 field experiments add up to

Bergquist, Nilsson and Schultz (2019) collected what was published between 1982 and 2019 in field experiments using normative messages on environmental behaviour: 91 experiments from 74 papers, 227,730 observed cases, 18 countries. Across all of it, norms work against control conditions at d = 0.32 (95 percent interval 0.28 to 0.37).

The figure that decides things for shop copy is a different one, though. The analysis separates two routes by which a norm reaches the recipient: spoken as a message — a sign, a letter, a sentence on a page — or read off the surroundings, such as a clean pavement or lights left off.

The difference between the two rows is clear (p < .001). Everything a shop can word sits in the lower row — a text is always a spoken message. And 0.18 sits below what usually counts as a small effect. The authors explain the gap by a spoken norm having a sender, which makes it recognisable as an attempt to influence (see reactance); a setting has none.

That fits the order of magnitude from the Opower programme. There the savings run at 2 to 5 percent; here the message form sits at d = 0.18. Both say the same thing: a cleanly built normative claim is a small, dependable gain — not a lever that overturns a conversion rate.

One qualification belongs with it directly. The studies using norms read off the surroundings mostly examined littering. The mode of conveyance and the behavioural domain therefore coincide, and only a single one of the included papers tested the two modes directly against each other. The gap is large enough to take seriously, but it is not cleanly isolated.

Where Goldstein’s towel sign sits in this collection is telling. Experiment 1 is listed at d = 0.19, whose interval begins at exactly 0.00, experiment 2 at d = 0.11. The two German repetitions by Bohner and Schlüter appear at d = −0.06 and d = −0.80, that is, with the sign reversed. The most-cited study in the field is among its weakest.

How specific should the reference group be?

Goldstein, Cialdini and Griskevicius (2008) tested several versions of the same towel sign in a single hotel. Here are the figures.

In the first experiment the descriptive norm beat the usual environmental appeal: 44.1 against 35.1 percent reuse (N = 433, Φ = 0.09). The Φ measures the strength of the association between two yes-no features and runs from 0 to 1; values around 0.1 count as small. In the second experiment five signs ran against each other. The version carrying the norm of that same room reached 49.3 percent, more than the three other normative versions combined (42.8 percent, N = 1,318, Φ = 0.06).

Three qualifications belong with that, and the usual retelling leaves them out.

The room version was tested against the average of the other three, not against the general guest norm on its own. That one sat at 44.0 percent, citizenship at 43.5, gender at 40.9 — and the three did not differ from one another (all p > 0.42). The head-to-head of “all guests” against “guests in this room” is not reported as a test of its own.

The effect is small. Φ = 0.06 means a difference of a good six percentage points. For a field experiment that is a respectable result, but it is not something that overturns a conversion rate.

It is one hotel, two experiments, one team. The second experiment does not replicate the first independently; it repeats it in the same place with the same authors.

A fourth finding sharpens the rule rather than limiting it. Respondents rated “guest in this room” the least important of the group memberships tested, markedly less important than gender or citizenship. Those two, the ones that mattered to people, did no better than the general guest norm. Closeness in this study means the same situation, not similar people.

An independent team has tested this rule, and it did not pass. Bohner and Schlüter (2014) repeated Goldstein’s setup in two German hotels, with 724 and 204 observed guests.

The same-room advantage reversed, and significantly so (78.0 against 85.6 percent, p = .017). In the second study the tendency went back in Goldstein’s direction but without significance — and that study, at 204 observations, was too small to find small effects at all (power of .21 to .30 by the authors’ own account).

More awkward still: no normative version beat the plain environmental appeal. In study 1 all four normative versions together came to 81.9 percent against 83.7 — no difference. In study 2 the environmental appeal was actually superior to the normative ones (93.3 against 76.5 percent, p = .039). And the starting point was different: with no message at all, 64.3 percent reused their towel — more than in any single condition in Goldstein.

Two further tests point the same way, both reported here at second hand through Bohner and Schlüter: Schultz and colleagues (2008) found a purely descriptive norm ineffective against a control group in the United States — only together with an injunctive norm did it work — and saw the general norm ahead of the room norm. Reese and colleagues (2014) in Austria and Switzerland found no advantage of any normative version over the environmental appeal.

For a German-speaking market a clear recommendation follows. There is no tested reason to prefer a normative figure over a clearly stated argument on the merits. The narrower reference group is not a lever but a risk.

The position is therefore not “unresearched” but “researched and practically without result” — a difference that matters in practice. What remains untested is how well a hotel finding carries over to a shop at all.

Practical cases

What we deliberately do not recommend

• Invented activity counts. “1,284 people are viewing this right now” without a real basis is misleading. That is not a design question.
• The minority norm out of honesty. “Few people use our product so far” is true and works against you. State something else that is verifiable.
• Social proof before the question of benefit. Someone who does not yet know what a product is for will not be convinced by a quantity. Social proof resolves uncertainty about the choice, not about the purpose.

Ethics check: nudge or sludge?

Limits and criticism