Social Proof, and the Boomerang
Telling people what most others do can move them toward the behaviour you want to stop.
Hollis, a payments tool for freelancers, wants more users on two-factor authentication. Marketing writes a banner: Most Hollis users haven't turned on 2FA yet. Protect your account. It is true, it is urgent, and it is precisely backwards. Two weeks in, adoption is flat and a support agent forwards a message from a customer asking whether 2FA is actually necessary, since apparently hardly anyone bothers.
We view a behavior as more correct in a given situation to the degree that we see others performing it.
When to cite the headcount
The descriptive norm cuts both ways
Cialdini's research team studied signage in Arizona's Petrified Forest National Park, where visitors steal petrified wood. A sign describing how many visitors remove wood increased theft compared with a control. A sign condemning the behaviour, without the prevalence claim, reduced it. Schultz, Nolan, Cialdini, Goldstein and Griskevicius reported a related pattern in 2007 with household energy use: households told they used less than average increased their consumption toward the mean. Adding an approving marker for the low users removed that rebound.
That is the boomerang. A descriptive norm (what people do) pulls behaviour toward the average from both directions. If the average is bad, you have just advertised it. If your best users are above average, you have given them permission to relax.
Two kinds of norm, and when each is safe
- Descriptive
- What people actually do. Safe only when the majority already does the thing you want. 68% of studios on Hollis invoice within 24 hours works. Most users haven't done X does not.
- Injunctive
- What people approve of. Works even when the behaviour is rare, because you are citing a standard rather than a headcount. Accounts handling client money are expected to use 2FA.
- Personalised comparison
- The strongest and the riskiest. Comparing a user to a relevant peer group moves behaviour hard, in both directions. Pair it with an injunctive signal so the above-average user is praised rather than released.
Proof that undermines you
- Most Hollis users haven't turned on 2FA yet.
- Join over 2,000 businesses
- 3 people are looking at this plan right now
- Rated 4.9 by our users
Proof that holds up
- Accounts that hold client funds are required to use 2FA from 1 June.
- Used by 340 UK design studios, including 12 in Bristol
- Typical setup time for a studio your size: 25 minutes
- 4.6 from 812 reviews on G2
Look at the second row. Over 2,000 businesses is a number chosen to sound large and to be unfalsifiable. 340 UK design studios is smaller and stronger, because it is specific, it is checkable, and it maps onto the reader. Cialdini's related point about liking and similarity applies: proof from people unlike the reader carries less weight than proof from people who look like them, and a big undifferentiated crowd is nobody in particular.
The fourth row is about credibility. A perfect rating reads as curated. A 4.6 with a visible sample size: How many observations a test needs before its result means anything. reads as a real distribution, and a reader who has ever written a review knows which of the two they can trust.
Worked example
Hollis rewrites the banner
New copy: From 1 June, accounts that receive client payments need 2FA. Setting it up takes about 2 minutes and works with any authenticator app. No prevalence claim at all. It is an injunctive norm plus a deadline plus an effort estimate, which is Fogg's ability lever. Adoption moves, and nobody learns how many of their peers skipped it.
Where honesty gets tested
Fake urgency counters, invented viewer counts and rounded-up customer totals all work in the short run, and they are the reason many readers now discount every social proof: People look at what others did to decide what is correct here. element on a page. If you run one of these, you are spending credibility that your product's real evidence needs later. Say the true number or say nothing.
- Before shipping any prevalence claim, check whether the majority is on your side. If not, switch to an injunctive frame.
- Segment the proof so the reader sees people like them, not a global total.
- Show the denominator. A rating without a sample size is decoration.
- If you compare a user to peers, decide in advance what the above-average user sees, because the default outcome is that they do less.
Quick check
Why did the Petrified Forest sign citing how many visitors take wood increase theft?
The takeaway
Never advertise that most people skip the behaviour you want; when the majority is against you, cite the standard rather than the headcount.
Try this tomorrow
Audit every social proof string in your product and marketing this week, and rewrite any that states or implies a minority is doing the desired behaviour.
Answer the check above, then bank the day.
Where this comes from
- Influence: The Psychology of Persuasion, Robert Cialdini
- The Constructive, Destructive, and Reconstructive Power of Social Norms (2007), Schultz, Nolan, Cialdini, Goldstein and Griskevicius
- Tiny Habits, BJ Fogg
User Psychology is one of six tracks. These lessons summarise and build on the work above, they do not reproduce it. Buy the books, they are better.