Your website says most clients choose the mid tier. Your renewal email says the average customer stays four years. Your invoice says ninety percent of clients pay within thirty days. Each of those is the same kind of statement, and the research says each one helps some recipients and harms others.
Key Takeaway
In a field experiment on household energy, a descriptive normative message detailing average neighbourhood usage produced either desirable energy savings or the undesirable boomerang effect, depending on whether households were already consuming at a low or high rate[1]. The descriptive norm increased consumption among low-consuming households[3]. Adding an injunctive message conveying social approval or disapproval eliminated the boomerang effect[1]. The authors offered this as an explanation for the mixed success of persuasive appeals based on social norms[4].
Our Grade For This Claim
Applying the scheme from the first article in this series.
Grade A. This is a large-scale field experiment measuring real behaviour rather than a laboratory task or a self-report, which is the strongest design category in the grading scheme. The authors themselves attribute the paper's influence to their methodology, which measured real behaviour in a large-scale field experiment, and note it is among the 30 most-cited articles across all APS publications[3].
Two qualifications, both ours.
We did not obtain the full text, so we report no effect sizes and no sample size, and we could not verify how one specific methodological concern was handled. That concern is set out in its own section below, along with why we think the design answers it.
And the setting is household energy consumption. The application of the finding to commercial messaging is our own extension and is not something the study tested.
A Note On Method
Everything here is verified to August 2026.
We obtained the published abstract from five independent sources reproducing it identically, including the publisher, PubMed and a university-hosted copy of the paper[1][2][4][6], plus the authors' own 2018 retrospective on the paper[3].
We did not obtain the full text of the 2007 paper and state no effect sizes, no sample size and no detail of the statistical treatment.
We did not obtain the hotel towel study, the focus theory paper, or the study on whether people know norms affect them, and cite each only as a secondary source describes it.
Where we simulate, extend or calculate anything, we say so and the work is ours.
This article reviews behavioural research. It is not marketing, advertising or legal advice. Claims made to consumers about what other consumers do carry obligations under consumer protection and competition law that this article does not address, and any such claim must in any event be true.
The Message Every Business Uses
The practice this article is about. This section is our own characterisation.
Social proof is close to universal in commercial communication. Most popular. Best seller. Chosen by 8 in 10 customers. Join 12,000 businesses. The average client saves X.
Two observations before testing it.
The reasoning behind it is sound and the underlying effect is real. People do use what others do as information about what to do, particularly where they are uncertain, and the research literature has documented this for decades.
But almost every commercial use of it shares a structural feature that turns out to matter enormously: the same message is sent to everybody. The website says it to every visitor. The email goes to the whole list. And the research below is about what happens when a normative message reaches someone who is already above the norm.
Two Kinds Of Norm
A distinction that predates the experiment and does all the work in it.
A source describes Cialdini and colleagues' focus theory of normative conduct as distinguishing descriptive norms, being what most people do, from injunctive norms, being what most people approve of, and as showing that descriptive norms are particularly effective in driving behaviour in public settings, attributing the theory to Cialdini and colleagues (1990)[5].
We did not obtain the 1990 paper and report this as that source describes it.
Three observations, ours.
The two are logically independent. What most people do and what most people approve of can point in opposite directions, and frequently do. Most people exceed the speed limit; most people approve of speed limits.
Commercial messaging is overwhelmingly descriptive. Most popular, best seller, chosen by most customers: these are all statements about what people do, not about what is approved.
And that is not an accident. A descriptive claim is verifiable from your own sales data, whereas an injunctive claim requires you to assert a value, which is a harder and more exposed thing for a business to do.
The Field Experiment
The study.
Schultz, Nolan, Cialdini, Goldstein and Griskevicius published The constructive, destructive, and reconstructive power of social norms in Psychological Science, 18(5), 429–434, in May 2007[2].
The problem they set out to solve is stated in the first line of the abstract: despite a long tradition of effectiveness in laboratory tests, normative messages have had mixed success in changing behavior in field contexts, with some studies showing boomerang effects[1].
Their design: a field experiment in which normative messages were used to promote household energy conservation[1].
Two observations, ours.
Note that the boomerang was already known before this study. The paper's contribution was not discovering it but testing a theoretical account of when it occurs, which is a more useful kind of contribution.
And note the phrase as predicted, which appears twice in the abstract. Both the boomerang and its remedy were predicted in advance rather than found by inspection. That distinction matters a great deal, for reasons the first article in this series set out about researcher degrees of freedom.
The Boomerang
The finding.
A descriptive normative message detailing average neighborhood usage produced either desirable energy savings or the undesirable boomerang effect, depending on whether households were already consuming at a low or high rate[1].
The authors' own later summary puts the direction beyond doubt: the descriptive norm was shown to increase consumption for low-consuming households, which they describe as this potential destructive effect of social norms[3].
A further account states it as providing households with information about neighbors' energy use reduced consumption among above-average users while elevating it among below-average users unless accompanied by an injunctive message[5].
Three consequences, ours.
The message did not merely fail for the light users. It moved them in the wrong direction, which is a different and worse outcome than no effect.
The effect on the two groups was opposite in sign, which means an aggregate measure would understate both. A programme evaluating only the average would see a modest saving and miss that it had made half its recipients worse.
And the people it harmed were the ones already doing the desirable thing. That is the part with the most direct commercial application, and it has its own section below.
The Fix
The second half of the finding, and the reason the paper is useful rather than merely alarming.
Also as predicted, adding an injunctive message (conveying social approval or disapproval) eliminated the boomerang effect[1].
The authors' retrospective describes the same result as the destructive effect being eliminated with the addition of an injunctive message of social approval for using less energy[3].
Three observations, ours.
The word is eliminated, not reduced. That is a strong claim and it is the authors' own.
The injunctive message was added, not substituted. The descriptive information stayed; approval was layered on top. So the remedy is not to stop telling people what others do.
And the logic is clean once stated. A descriptive norm tells a light user they are unusual. An injunctive norm tells them they are unusual in a good way. Without the second, being told you are below average is simply information that you have room to move.
An Objection Worth Taking Seriously
A methodological concern we raised against this study before writing about it. This section is our own.
Consider the reported pattern in the abstract: high users went down, low users went up.
That is also the exact signature of regression to the mean, which occurs in any repeated measurement containing noise, with no treatment whatsoever. Households measured as high in one period are partly high because of transient factors, and will tend to be lower next period regardless of what anybody sends them.
We ran a simulation to see how strong that artefact is. Generating households with a stable underlying consumption level plus period-to-period noise, and applying no treatment at all:
Of households above average in the first period, roughly 66 percent went down in the second.
Of households below average in the first period, roughly 66 percent went up.
These are our own simulated figures under assumptions we chose, not anything from the study, and they are illustrative of a mechanism rather than an estimate of anything.
The point they establish is that the raw high-down, low-up pattern proves nothing on its own. Any competent reader should want to know how the study handled this, and we could not verify how it did, because we did not obtain the full text.
Why The Design Answers It
Our own resolution of the objection, offered as reasoning.
The study's key finding is not the raw pattern. It is the contrast between two message conditions: descriptive alone, and descriptive plus injunctive.
Regression to the mean would apply equally to both conditions. It is a property of measurement, not of messages, and there is no mechanism by which it would affect households receiving a smiley face differently from households receiving the same usage figure without one.
So the fact that the boomerang was eliminated in one condition and not the other is the part regression cannot explain, and that is the finding the paper rests on.
Two consequences.
The comparative result is robust to the objection, on our reading, even though the raw descriptive pattern is not.
And that is a reason to be careful about how this finding is quoted. The defensible claim is that adding an injunctive component changes the outcome for low users. The claim that a descriptive message causes a precisely quantified increase is less well supported by anything we obtained.
We include this section because the objection is obvious to anyone who has looked at repeated-measures data, and a reader who spotted it and found no acknowledgment would be right to distrust the rest.
Why A Norm Cuts Both Ways
The underlying logic, stated plainly. This section is our own analysis.
A descriptive norm communicates a location. It says: here is where most people are.
That single piece of information does two different things depending on where the recipient already stands.
To someone above the norm on an undesirable behaviour, it says you are doing more than you need to. That is a reason to reduce.
To someone below the norm, it says you are doing less than everyone else. That is a reason to increase.
Two consequences.
A norm is a magnet, not an arrow. It pulls people toward the middle from both directions rather than pushing everyone one way, and the direction of pull is determined by the recipient rather than by the sender.
Which means a normative message can only ever move half your audience the way you want, unless you either segment the audience or add something that supplies direction. The injunctive component is what supplies direction.
Who Are You Actually Talking To
The commercial translation, and it is ours.
Any normative claim on a website, in an email or on an invoice reaches a mixed audience. Some recipients are above the stated norm and some below.
Three questions follow for any such message.
Which side of the norm is the desirable side? For energy, low is desirable. For most commercial behaviours, the desirable side is high: buying more, staying longer, paying sooner.
Which recipients are already on the desirable side? Those are the ones the message will pull the wrong way.
And how valuable are those recipients? In a commercial setting, the customers already exceeding the norm on spend, tenure or payment speed are frequently the most valuable ones you have.
That produces an uncomfortable conclusion, developed two sections below: the standard use of social proof is aimed most damagingly at your best customers.
Where This Bites In A Business
Specific cases, all our own extension of a finding tested only on household energy.
Tier messaging. "Most clients choose the standard package" is read by a client on the premium package as information that they are paying more than they need to.
Average spend. "Our typical client invests $X annually" tells the client investing 2X that they are an outlier.
Tenure. "The average client stays with us four years" tells a seven-year client they have already had more than most.
Engagement. "Most users log in twice a month" tells your daily user they are using it unusually heavily.
And compliance or deadline messaging, which the next section takes on its own because it is the sharpest case.
Two cautions, ours. None of these has been tested; they are applications of a mechanism to settings the study did not examine. And the effects in a commercial setting may be far smaller than in a repeated household behaviour with a monthly bill attached.
The Payment Terms Example
One case worked through, because it is the one most likely to be actionable for a small business. This is our own reasoning.
A firm with a late payment problem puts a line on its invoices: "Nine out of ten clients pay within thirty days."
The intent is clear and the statement is true. Consider its two audiences.
The late payer learns they are unusual. On this literature, that is the pull you wanted.
The client who habitually pays in ten days learns that thirty is normal and that nine in ten of their peers take longer than they do. On this mechanism, that is an invitation to relax.
Two observations.
The firm's working capital depends disproportionately on the early payers, and they are the group the message pulls the wrong way.
And the effect would be invisible in the aggregate. Average days to payment might improve while your best payers drift out, and you would record that as a success. This is the same measurement asymmetry that has appeared in six previous articles in this series.
We are not claiming this happens. We are claiming it is the prediction the mechanism makes, and that a firm using such a line has no way of knowing which effect dominates without segmenting.
You Are Aiming At Your Best Customers
The general form of the problem, stated once. This section is our own.
In almost every commercial application, the norm sits below your best customers by definition, because a mean is dragged down by the majority.
Three consequences follow.
A descriptive norm is reassuring to your weakest customers and deflating to your strongest, which is the opposite of how most businesses would allocate a persuasive effort if asked.
The strongest customers are also the ones with the most room to fall. A client at twice the average spend has considerably further to regress than a client at half of it has to rise.
And because those customers are fewer in number but larger in value, an aggregate metric weighted by headcount will not detect the damage.
We offer this as reasoning from the mechanism rather than as a finding. But it is a reasoning chain a business can check against its own data, which the first article in this series argued is better evidence for you than any published study.
What An Injunctive Component Looks Like
The practical remedy. The underlying finding is the study's; the commercial translation is ours.
The study added a message conveying social approval or disapproval[1], described elsewhere as social approval for using less energy[3].
Translated to a commercial message, the descriptive claim needs a companion that says which direction is the approved one.
"Most clients choose the standard package" becomes a statement that also says something about what the premium client is doing right.
"Nine out of ten clients pay within thirty days" becomes a statement that also thanks and commends the ones who pay sooner.
Three cautions, and they are firm.
The commercial versions above are untested illustrations of a principle, not validated copy, and we did not obtain the study's actual message wording.
An injunctive claim asserts a value, which is a more exposed thing for a business to say than a fact about its sales, and it should be one the business is willing to stand behind.
And every claim, descriptive or injunctive, must be true. Consumer protection and competition law govern representations to consumers, this article does not address them, and nothing here licenses a claim that is not accurate.
People Do Not Know This Affects Them
A related finding, reported briefly because our sourcing is second-hand.
A source states that Nolan and colleagues (2008) revealed that people systematically underestimate the influence of social norms on their own behavior while overestimating the role of other factors, and our source truncates at that point[5].
We did not obtain that study and report only what this description says.
Two observations, ours, offered cautiously.
If that holds, it means self-report is not a usable check on whether a normative message worked. Asking customers whether a "most popular" label influenced them would produce an answer, and the answer would understate.
Which parallels the anchoring article in this series, where experts insisted a number had not affected them while being demonstrably affected by it. Two unrelated literatures, the same failure of introspection about the same class of influence.
What This Says About The Nudge Dispute
Connecting to the ninth article in this series. This section is our own.
That article described a public dispute over whether nudging works at all, with a meta-analysis of 447 experiments concluding it does and a bias-corrected reanalysis concluding no evidence remains.
This study is useful context for that dispute, for three reasons.
It is a single well-designed field experiment measuring real behaviour, which is exactly the kind of evidence that survives the heterogeneity critique. It is not an average over hundreds of unrelated interventions.
It predicted its results in advance, which the abstract states twice, and preregistration-adjacent practice of that kind is what the replication literature identifies as protective.
And it establishes both an effect and its failure mode. A literature that can say when an intervention backfires is in a considerably better position than one that can only report an average.
Our own view: this is roughly what a usable behavioural finding looks like, and it is a better model than the pooled meta-analysis that the ninth article described being fought over.
The Segmentation Requirement
The practical constraint nobody mentions. This section is our own reasoning.
The study could distinguish high from low users because it had their consumption data. Every household had a meter.
Three consequences for a business.
Applying this properly requires knowing where each recipient sits relative to the norm, which is a data question before it is a messaging question.
For most businesses that data exists for existing customers, in the ledger, and does not exist for website visitors, who are the audience most social proof is aimed at.
Which suggests a division: descriptive social proof is safest where the audience is uniformly below the norm, which describes a prospect who is not yet a customer at all, and most dangerous in communications to an existing base whose members sit on both sides.
We offer that as a reasoned rule rather than a tested one.
What To Do
Work out which side of your norm is the desirable side. For most commercial behaviours it is the high side, which means your best customers are the ones a descriptive norm pulls the wrong way.
Add an injunctive component to any normative claim. The study found that adding a message conveying approval or disapproval eliminated the boomerang, and the descriptive information stayed in place.
Do not send a single normative message to a mixed base. A norm is a magnet, not an arrow, and it pulls from both directions at once.
Be most careful on invoices and renewal notices. Those go to existing customers whose position relative to the norm you already know from your own records.
Treat prospect-facing and customer-facing social proof differently. A prospect at zero is below every norm you could state; an existing base is not.
Do not evaluate on the aggregate. Opposite effects on two segments will partly cancel, and the segment you damaged is likely the smaller and more valuable one.
Do not ask customers whether it influenced them. The one study we found on this reports people systematically underestimating how much norms affect them.
Check it against your own ledger. You have payment dates, tenures and spend by customer. Whether your best payers drifted after you added a line to the invoice is answerable from data you already hold.
Make sure every claim is true. Representations to consumers are governed by law this article does not address, and none of the above licenses an inaccurate statement.
The Limits Of This Analysis
Several caveats matter. This article reviews behavioural research and is not marketing, advertising or legal advice; representations about what other customers do carry obligations under consumer protection and competition law that are not addressed here, and any such claim must be true. Everything is verified to August 2026. We did not obtain the full text of the 2007 study; we rely on its published abstract, reproduced identically by five independent sources, and on the authors' own 2018 retrospective. We state no effect size, no sample size and no detail of the statistical treatment, and we did not obtain the actual wording of the messages used. We could not verify how the study handled regression to the mean, and we set out both the objection and our own reasoning for why the comparative design appears to answer it; that reasoning is ours and is not something we confirmed from the paper. Our simulated figures on regression to the mean are our own, produced under assumptions we chose, illustrate a mechanism rather than estimate anything, and appear in no publication. We did not obtain the focus theory paper, the hotel study, the study on people underestimating normative influence, or any subsequent replication, and cite each only as a secondary source describes it. The study concerns household energy consumption. Every commercial application in this article, including the tier, spend, tenure, engagement and payment terms examples, is our own extension to settings the study did not test, and effects in those settings may differ in size or direction. Roughly nineteen years of subsequent literature was not reviewed.
Frequently Asked Questions
Does social proof work?
What is the boomerang effect?
What is the fix?
Could this just be regression to the mean?
Where is this most dangerous in a business?
How would I know if it were happening to me?
References
- Schultz, P. W., Nolan, J. M., Cialdini, R. B., Goldstein, N. J., & Griskevicius, V. (2007). The constructive, destructive, and reconstructive power of social norms. Psychological Science, 18(5), 429–434. DOI 10.1111/j.1467-9280.2007.01917.x, published abstract via PubMed, on normative messages having had mixed success in changing behavior in field contexts despite a long tradition of effectiveness in laboratory tests, with some studies showing boomerang effects; on the authors conducting a field experiment in which normative messages were used to promote household energy conservation; on a descriptive normative message detailing average neighborhood usage producing either desirable energy savings or the undesirable boomerang effect depending on whether households were already consuming at a low or high rate; and on adding an injunctive message conveying social approval or disapproval eliminating the boomerang effect. Note: we obtained the published abstract only and state no effect size, sample size or message wording. pubmed.ncbi.nlm.nih.gov
- Publisher record for Schultz and colleagues (2007), Psychological Science, 18(5), 429–434, confirming authorship, journal, volume, issue, pages and date, and reproducing the abstract identically. Note: the publisher's own record, used to confirm the citation and corroborate the abstract. journals.sagepub.com
- Schultz, P. W., Nolan, J. M., Cialdini, R. B., Goldstein, N. J., & Griskevicius, V. (2018). The Constructive, Destructive, and Reconstructive Power of Social Norms: Reprise. Perspectives on Psychological Science, on the 2007 field experiment having shown that social norms could be leveraged to promote residential energy conservation but that the descriptive norm was shown to increase consumption for low-consuming households; on this potential destructive effect having been eliminated with the addition of an injunctive message of social approval for using less energy; and on the article being among the 30 most-cited articles across all APS publications, which the authors attribute to a methodology that measured real behavior in a large-scale field experiment. Note: the original authors' own retrospective, which is not a neutral source but is authoritative as to what they claim their study showed. scranton.edu
- University-hosted copy of Schultz and colleagues (2007), reproducing the abstract and confirming the authors' institutional affiliations at the time of publication, and stating that the results offer an explanation for the mixed success of persuasive appeals based on social norms and suggest how such appeals should be properly crafted. Note: a hosted PDF; we obtained the first page only. assets.csom.umn.edu
- ResearchGate record for Schultz and colleagues (2007) with associated third-party commentary, on Cialdini and colleagues' (1990) focus theory of normative conduct distinguishing descriptive norms, being what most people do, from injunctive norms, being what most people approve of, and showing descriptive norms to be particularly effective in public settings; on Schultz and colleagues (2007) having demonstrated that providing households with information about neighbors' energy use reduced consumption among above-average users while elevating it among below-average users unless accompanied by an injunctive message; and on Nolan and colleagues (2008) having revealed that people systematically underestimate the influence of social norms on their own behavior while overestimating the role of other factors. Note: a publisher record with third-party commentary, not peer-reviewed; we did not obtain the 1990 focus theory paper or the 2008 study and report both only as this source describes them, with the latter truncated. researchgate.net
- Experts@Minnesota institutional repository record for Schultz and colleagues (2007), confirming the citation details and reproducing the abstract. Note: an institutional repository record, used as a further independent confirmation of the citation and abstract text. experts.umn.edu
This article reviews behavioural research and is not marketing, advertising or legal advice. Representations to consumers about what other consumers do are governed by law not addressed here and must in any event be true. The 2007 study was not obtained in full; no effect size, sample size or message wording is stated. How the study handled regression to the mean could not be verified, and the reasoning offered for why its design answers that objection is the authors' own. The simulated figures illustrating regression to the mean are the authors' own and appear in no publication. The study concerns household energy; every commercial application here is an untested extension.