Social proof in marketing, and what it proves

Most social proof on a small business site does not prove anything

Social proof is one of the better-evidenced ideas in this series. The way most small businesses use it is not.

A row of 5-star ratings with no names attached, a wall of client logos, an unsourced number of happy customers: none of these gives a buyer anything they can check. My position is that the useful work is making a small amount of evidence specific and verifiable, rather than making a large amount of it visible. A buyer deciding on a considered service is not counting your endorsements. They are looking for one person like them who can be asked.

Where the idea comes from

Robert Cialdini named the principle in 'Influence: The Psychology of Persuasion' in 1984, alongside reciprocity, commitment and consistency, liking, authority and scarcity. His formulation is that we decide what is correct by finding out what other people think is correct, and that we lean on this most when we are uncertain and when the other people resemble us.

Those 2 conditions do most of the work and get dropped from the marketing version. Uncertainty and similarity are the moderators. Proof from someone in a different situation, facing a different problem, at a different size of business, is not doing what you think it is doing.

The idea sits on older conformity research, Sherif on ambiguous judgements in the 1930s and Asch on visible group pressure in the 1950s. That lineage matters because it sets the boundary: the effect is strongest where the right answer is unclear, which describes buying a professional service and does not describe buying a familiar product at a known price.

What the field evidence supports, and where it stops

Goldstein, Cialdini and Griskevicius ran the most quoted field test in A Room with a Viewpoint, Journal of Consumer Research 2008. Hotel signs telling guests that most other guests reuse their towels outperformed the standard environmental appeal, and signs naming the guests in that same room did better still.

The caveat belongs with the finding. A German replication published in PLOS One in 2014 found both messages raised reuse against a no-message baseline, but the descriptive norm message was no better than the standard one. Take the mechanism as supported and the effect size as unsettled.

The more useful result for a business is the one about failure. Schultz, Nolan, Cialdini, Goldstein and Griskevicius, in The Constructive, Destructive, and Reconstructive Power of Social Norms, Psychological Science 2007, sent households their neighbourhood's average energy use. Heavy users cut consumption. Light users increased it, moving up towards the average. Telling people what most people do pulls behaviour towards the average from both directions. Adding an approving note to the below-average households removed that effect.

The commercial version of that problem is common and rarely noticed. If you tell a prospective client that most of your clients spend £2,000 a month, you have told the ones who were going to spend more that they are overpaying.

There is a third finding worth carrying. In Normative Social Influence is Underdetected, 2008, the same group found descriptive norms predicted behaviour better than any other belief measured, while respondents rated those norms as the least important thing affecting them. People are influenced by what others do and do not believe they are. That is a reason to be careful about asking buyers directly what persuaded them.

The dual-process framing this article used to rely on

An earlier version of this page explained social proof as something that works on System 1, the fast automatic system, while System 2 sits out the decision. I have removed that framing.

The 2-system model is a teaching device rather than a description of 2 mechanisms in the head. Kahneman called the systems fictitious characters, and a critical review of dual-process theories sets out how much weaker the general account is than its popular version. I go through this in the power of the autopilot and in the science of consumer behaviour. Fast impressions and slower reasoning interact throughout a decision. There is no separate channel that reviews reach and reasoning does not.

The same page claimed that reducing the number of choices reduces decision fatigue and improves mental well-being. The first half is contested and the second was never supported by anything. A meta-analysis of 99 observations involving 7,202 participants found choice overload is heavily moderated by task difficulty, preference uncertainty and the buyer's goal, with a mean effect close to zero. Fewer options do not automatically convert better, and no marketing decision has been shown to affect anyone's mental health.

Ratings do not behave the way most sites assume

Northwestern's Medill Spiegel Research Center published How Online Reviews Influence Sales in 2017. There are 2 findings worth holding.

The first few reviews carry most of the value. Purchase likelihood for a product with 5 reviews was 270% higher than for the same product with none, and the gain per review falls away sharply after that. Getting from 0 to 5 reviews matters more than getting from 50 to 100.

Purchase likelihood peaks in the 4.0 to 4.7 range and declines as the average approaches 5.0. A perfect score reads as unverified rather than excellent.

The limits are real. This is observational ecommerce data from a US retail panel, not a controlled experiment, and it covers products rather than services. I use it as a reason to publish the mediocre reviews alongside the good ones, not as a number to quote at a client.

Grade the proof you already have

This is the part I would spend an afternoon on. Take everything on your site that is meant to reassure a buyer and put it through this table. The column that decides the answer is the third one.

Type of proof	What it proves	How a buyer verifies it	The stronger version
Star rating on your own site	That you chose to display a number	Nothing they can check	The same rating on a platform you do not control, with the low scores still visible
Testimonial with a first name only	That someone said something, possibly	Nothing they can check	Full name, job title, organisation, and a link to that person
Client logo wall	That you did some work for those organisations, at some point, of some size	Nothing they can check	2 logos with a sentence each saying what the work was and when
Follower or subscriber count	That an audience exists	Visible, but tells them nothing about the work	Drop it. Audience size is not evidence of competence
Case study with no numbers	That a project happened	Only the narrative	The starting position, what changed, and what you cannot attribute to your work
Named referee a buyer can phone	That a real client will stake their time on you	Directly, by phoning them	2 referees at the buyer's size and in their situation
Third-party review platform	That reviews came from somewhere you cannot edit	Independently, including the bad ones	Reply to the critical reviews in public

The pattern the table exposes is that most social proof on a small business site is assertion in the shape of evidence. Fixing it usually means having less of it.

I ran the table over a business I work with

PCLA are independent loss assessors working in Northern Ireland and Scotland, and I work with them. Putting their proof through the table above is more useful than a hypothetical, partly because the result is mixed.

The strongest item passes the third column outright. The site carries 276 5-star Google reviews as I write this. That is stronger than the same praise reproduced as testimonials on the site itself, because a buyer can leave the website, look at the reviews on Google and see who wrote them. We cannot edit them, and that is the whole of their value. The count also moves, which is worth saying. It rises most months, so any figure of this kind is a snapshot and should be written as one.

The published case studies are the more interesting case, because they sit between the 2 grades. They give the starting position and what changed: an insurer offer of £6,000 that became a £73,000 settlement with accommodation paid, in Belfast. A £700 offer that became £10,900 in Mansewood, Glasgow. A £3,000 storm damage offer in Perthshire that became £15,000. Describing what changed is more useful to a reader than a claim about maximising settlements, because it is specific enough to be wrong.

They are still assertions on our own page. A prospective client cannot independently confirm any of them, and I would not present them as though they could. There is a second limit that matters more, and it is the one most case studies skip. Those figures do not establish that PCLA alone caused the difference. Claims develop, damage that was not visible at first inspection gets found, and insurers revise their own positions. Some of that would have happened anyway.

So the honest grading is that the Google reviews are independently checkable and the case figures are not. If I were choosing what to strengthen first, it would not be adding more case studies. It would be finding out how much identifying detail a client will let us publish alongside one of them.

One further observation, and I am marking it as the weakest thing in this article. Warm enquiries that stall for weeks, with no objection stated, very often end with a request to speak to somebody. Nothing about that was counted, no comparison group exists, and I cannot tell you how often it happens across more than 20 years in communications and marketing. I offer it as the reason I would rather have 2 clients willing to take a phone call than 30 anonymous testimonials.

If you have no social proof yet

This comes up constantly and most writing on the topic ignores it.

Do not manufacture it. Borrow adjacent credibility instead and say plainly what it is. The work you did before you started this business is evidence. So is a named person who will vouch for how you work rather than for this specific service. So is showing the method itself in enough detail that a buyer can judge it, which is the closest thing to proof available before you have clients. Publishing the way you approach a problem is slower than collecting testimonials and it does not depend on anyone else. What you are standing in for is the buyer's difficulty in judging quality before they commit, which is one of the conditions that makes a decision feel risky. The psychology of risk sets out the others.

Where this becomes unlawful in the UK

The unfair commercial practices provisions of the Digital Markets, Competition and Consumers Act 2024 apply to practices from 6 April 2025. Schedule 20 lists practices that are unfair in all circumstances, with no need to show a consumer was harmed. Submitting or commissioning a fake review is on that list, as is publishing reviews without taking reasonable and proportionate steps to check whether they are genuine, and failing to disclose that a review was incentivised. The CMA has published its fake reviews guidance, CMA208.

There are 3 consequences for a small business. Asking clients for a review in exchange for a discount requires disclosure. Publishing only your positive reviews while suppressing the rest is a misleading omission. Writing a review of your own business is prohibited outright, whatever it says. I am describing the shape of the rules rather than giving legal advice, and I cover the same Act on urgency and scarcity in the role of emotions in purchase decisions and on presentation of choices in choice architecture.

How you would know it was working

You cannot ask, given the underdetection finding above. Here are 3 substitutes that cost nothing.

Count how many enquiries reach the reference request. If that stage is disappearing, the proof on the site is doing its job earlier.

Watch which case study gets mentioned unprompted on a first call. That tells you which buyer situation your evidence matches, which is usually narrower than you assumed.

Note how long a warm enquiry sits before someone comes back. Long silences with no objection are the signature of a buyer trying to justify the choice to somebody else, which is a proof problem in a different form. I set out that same pattern in designing effective decision interfaces.

What to do with this

Run the table over your own site this week and delete anything that fails the third column. The site will look emptier and say more.

Then find 2 clients who will take a phone call, ask them, and say on the page that references are available at that level of specificity. That single line does more than the wall of logos it replaces.