Social proof examples: cheap signals compared with costly ones

A number of other people did something. That is the whole of the evidence. Everything useful comes from working out who those people were, what they were choosing between, and whether anyone could have stopped them appearing.

What to take away

Signals that cost something to produce

These are the strongest, because appearing in them required a person to give up money, time or a reputation.

1. The busy restaurant on a weeknight. Covers: enough people found it worth going out for on an ordinary night. Does not cover: whether you will like the food. It is a genuine signal and it is measuring a different thing than the one you are asking about.

2. Repeat customers rather than total customers. Covers: people came back after knowing what they were buying. This is the single most informative retail number, and it is the one least often published.

3. A tradesman with a waiting list. Covers: demand exceeds supply at the current price. Does not cover: quality relative to someone with free slots, who may have raised capacity rather than lost customers.

4. A book still in print after forty years. Covers: successive generations of buyers kept choosing it without a marketing budget. Survival over that long is expensive to fake.

5. Professionals who use it themselves. Covers: people with information you lack, spending their own money. Ask whether they were given it free, because that turns the signal into something else entirely.

6. The queue outside in bad weather. Covers: people paid a real cost to be there. Also the case most often staged, which is why it appears here and again below.

Signals that are cheap to produce

Cheap does not mean worthless. It means the signal needs a second check before it carries weight.

7. Follower counts. Covers: an account accumulated followers by some route. Does not cover: which route, and the routes differ enormously in what they imply.

8. Star ratings with a small number of votes. Covers: very little. A handful of ratings is compatible with almost any underlying reality, and the average is doing more work than it can support.

9. "Most popular" on a pricing page. Covers: nothing verifiable. It may be a count, and it may be a label chosen by whoever set the prices, and from outside these look identical.

10. Testimonials on a seller's own site. Covers: that these people said these things, probably. Does not cover: what the people who were not selected said. The Federal Trade Commission's guidance on endorsements, influencers, and reviews sets out what a business is expected to disclose here, and reading it tells you what the absence of a disclosure means.

11. Download or install numbers. Covers: people acquired it. Does not cover: whether they kept it, which is the question you actually have.

12. "Join thousands of others." Covers: nothing at all. There is no verifiable claim in the sentence, and its function is tone rather than information.

Signals that were manufactured

13. The seeded queue. People are paid to stand in line, or the door is opened slowly so that a line forms. The visible fact is true and the inference it produces is false, which is the defining shape of the whole category.

14. Review farms and coordinated posting. A block of positive reviews appearing in a short window, in a similar register, from accounts with no other history. Design tactics of this family are cataloged in the Federal Trade Commission's staff report Bringing Dark Patterns to Light.

15. The counter that only goes up. A live count of people viewing, buying, or joining, generated rather than measured. Nothing is being reported; a number is being displayed.

The reading each type supports

Notice that even the top row stops short of your question. A crowd tells you about the crowd. Turning that into a statement about your own decision requires you to believe those people wanted what you want, and that step is yours to make or refuse.

A flock turns together because each bird is watching a few neighbors. It is an efficient system and it has no opinion about where the flock should go, which is roughly the position you are in when you follow a crowd signal without asking how it was produced.

Using the cases

Take any signal you are about to act on and place it in one of the three groups. If it belongs in the first, ask what the crowd was choosing between. If it belongs in the second, look for one costly signal that agrees with it. If it belongs in the third, treat the number as decoration.

Then ask the question that runs through this whole site: if the person showing you the number explained exactly how it was produced, would it still move you? An honest count survives that.

The mechanism and the five questions to ask of any published figure are set out under social proof. The version that runs on wanting to fit in rather than on evidence is normative influence.

Where the crowd is being treated as knowing something you do not, the channel is informational influence, and the general framework is social influence basics.

Common questions

Which of the fifteen is the most useful in practice?

Repeat purchase. It is the only common signal where the person had full information about the product before choosing again, and that is exactly the condition your own decision is missing.

How do I judge a rating with very few votes?

Treat it as a single anecdote rather than an average, and look for whatever costly signal exists alongside it. A small sample does not become more informative by being expressed as a decimal.

Are all seeded signals dishonest?

The ones that create a false impression are, whatever the intention. A visibly sponsored placement is a different thing, because the person can see what they are looking at and weigh it.

What if a crowd signal is all I have?

Then act on it and know what you are doing. A weak signal used knowingly is fine. The failure is treating a weak signal as a strong one because it was presented confidently.