A 61-million-person experiment on whether social proof changes behavior

You see a post. It has 10,000 likes. You wonder if the number made you stop. You wonder if it would work on your own posts. A bigger number. More proof that others cared. Would that make the next person care too?

Most writing about social proof is theory. It talks about why it should work. This post is about one experiment that measured whether it does. On 61 million people. All at once. The researchers did not ask people what they might do. They changed what people saw. Then they counted what people did.

The finding is the hardest evidence we have. Seeing that specific friends had already acted made people more likely to act themselves. Not just seeing a number. Seeing which friends. And the effect flowed through close ties. Not weak ones.

One thing before the rest. This measured voting. It did not measure Instagram reach. It did not measure engagement on posts. Reach is a harder problem, and the research on it says nobody can predict which post takes off. This study measured one behavior. On one platform. On one day. This post will say that plainly every time it matters.

Every source below is linked and dated. One section names what this experiment cannot tell you.

The experiment that is hard to argue with

November 2, 2010. Election day in the United States. About 61 million people logged into Facebook. A banner appeared at the top of their feed.

The researchers split those 61 million people into three groups. At random. Like a coin flip. That is what makes this hard to argue with. You do not have to wonder whether the groups were different to start with. Random assignment settles it.

Group one saw a banner. It showed a button that said "I Voted". It showed a counter. The counter said how many Facebook users had already voted. And it showed photos. Photos of your friends who had clicked the button.

Group two saw the same banner. The same button. The same counter. No photos.

Group three saw nothing.

Group one had about 58.8 million people. Groups two and three had about 600,000 each.

Then the researchers checked the voting records. Real voting records. From real polling places. They matched Facebook accounts to those records. The algorithm masked who was who once the match was made. So Facebook could not see which specific person voted. But the researchers could count how many did.

The results came from Bond, Fariss, Jones, Kramer, Marlow, Settle and Fowler. They published it in Nature in 2012. The paper is titled "A 61-million-person experiment in social influence and political mobilization". Nature is a peer-reviewed journal. We could not access the paper itself. It sits behind a paywall. What follows comes from secondary reporting. The Smithsonian Magazine covered the study when it came out. That is the source we could read.

Seeing your friends made the difference. Seeing a number did not.

Group one voted more than group three. The group that saw friend photos voted 0.39% more than the group that saw nothing. That is 0.39 percentage points. Not 0.39 times as much. The difference is small in percentage terms. The number of people is what made it large. About 60,000 more people voted because of the message they saw.

Here is the part that cuts through everything else. Group two saw the counter. It saw the button. It saw how many people had voted. It did not see the friend photos. That group voted at the same rate as the control group. The informational message did nothing.

The Smithsonian article quotes James Fowler. He is one of the authors. His words:

Social influence made all the difference in political mobilization. It's not the 'I Voted' button... that gets out the vote. It's the person attached to it. — James Fowler, quoted in Smithsonian Magazine, 26 September 2012

That sentence is the finding. Not the count. Not the button. The photos of specific people you know.

The effect traveled through close friends, not everyone

The experiment also measured something indirect. Some people in group three saw nothing. But their friends saw the banner. Did that change how group three voted? It did. If your close friends saw the social message, you were more likely to vote. Even if you never saw the banner yourself.

The study defined close friends by Facebook behavior. How often you tagged each other in photos. How often you sent messages. The indirect effect ran through those ties. Not through weak ties. Not through people you barely interacted with.

The researchers estimated the indirect effect at about 280,000 additional votes. That is much larger than the direct effect. The Smithsonian article says most of that came from offline conversations. People saw their friends had voted. Then they talked about it. In person. That is not the same as a counter going up on a screen.

So the total effect was about 340,000 additional voters. 60,000 directly. 280,000 indirectly.

This is about voting. It is not about Instagram reach.

We need to be exact about what this experiment did and did not measure. It measured one thing. Turning out to vote. On one day. On one platform. In one country.

It did not measure Instagram reach. It did not measure likes, shares, comments or follows. It did not measure purchases. It did not measure clicks on posts. It did not measure anything on any platform that recommends content.

The mechanism is plausible. Seeing that real people engaged with a post could make the next person more likely to engage. But no published experiment we could find proves that mechanism inside Instagram's ranking system. Or inside any other system that recommends posts.

So what can you take from this? The experiment proves that seeing specific people you know had acted makes you more likely to act. It proves that effect is real. It proves it is not just in people's heads. And it proves the effect flows through close connections.

What it does not prove is the size of any such effect on a different behavior. Or on a platform where an algorithm decides who sees what. That is a different question. This experiment does not answer it.

What makes this experiment harder to argue with than the others

Other studies have looked at social influence. Three get cited often. Salganik, Dodds and Watts built a fake music market in 2006. Muchnik, Aral and Taylor ran a test on a news site in 2013. Van de Rijt and colleagues gave random success to people on four websites in 2014.

All three found that early signals mattered. A fake upvote stuck. A fake download count changed later downloads. But none of them ran on a platform like Instagram. Or Facebook. Or anywhere that uses a model to decide what to show you.

The Bond experiment is different in three ways.

First, it was huge. 61 million people. That is not a sample. That is a population. The Salganik study had 14,341 participants. Van de Rijt had a few hundred per site. Bond had millions. The size makes the result stable.

Second, it measured real behavior. Not downloads of songs by unknown bands. Not ratings on comments. Real voting. At real polling places. Verified against public records.

Third, the groups were truly random. Some large studies just compare groups after the fact. This one assigned people to groups by chance. That settles whether the groups were the same to start with. They were.

What makes it less useful for Instagram is simple. You cannot see how many people engaged with a post before you did. The count does not show up on your feed. The signal runs through the ranking model instead. So the path from one person's engagement to the next person's exposure is hidden. It exists. But it is not the same as a visible counter with friend photos.

What this tells you about real engagement versus fake

The experiment makes one thing clear. Generic social proof did not work. The informational message had no effect. Seeing a big number did nothing. What worked was seeing that specific people you know had acted.

That has a consequence for anyone buying engagement. Fake engagement is a number. It is not a person you know. The Bond experiment says the number alone does nothing. The person does.

The Bond experiment does not prove that engagement from real accounts moves anything on Instagram. No experiment we found does. But the voting study says where a signal comes from matters. The person attached to it did the work. The count did not.

Fake followers are not people anyone knows. Fake likes are not from close friends. If the Bond finding holds elsewhere, fake engagement should do less than real engagement from real accounts. That is reasoning from one study about voting. It is not a controlled test.

What we could not determine

Five things.

Whether the Bond finding applies to Instagram reach. The experiment measured voting. We found no published experiment that measures whether early engagement from real people increases reach inside Instagram's recommendation system. The mechanism is plausible. The proof does not exist.

The size of any such effect on Instagram. Even if the mechanism works, the Bond study gives no number for it. 0.39 percentage points applied to voting. That number does not transfer.

Whether weak ties matter on Instagram. The Bond study says weak ties carried no effect for voting. Instagram shows your posts to people who do not follow you. Those are weak ties. Or no ties. Whether engagement from strangers helps your reach is unproven.

The exact definition of a close friend in the Bond study. We know it was based on Facebook interactions. The Smithsonian article does not give the threshold. We could not read the paper itself to check.

The authors' stated limitations. Every good study names what it cannot prove. We could not read the paper. So we could not quote the authors' own caveats. The limitation above is ours. Not theirs.

How we did this

Scope: this post reads one secondary source about a study we could not access directly. The Bond et al. paper is paywalled in Nature. We tried multiple URLs. None gave us the text. What we report comes from the Smithsonian Magazine article covering the study. That is a secondary source. We treat it as such.

Limits: we did not read the original paper. So we did not verify the numbers against the source. We did not quote the authors' own words except where the Smithsonian article quoted them. And we did not read the authors' stated limitations. Everything in the "what we could not determine" section is our inference. Not the authors' caveat.

Sources

1. How a Facebook experiment increased real-world election turnout, 26 September 2012. Smithsonian Magazine. Secondary coverage of the Bond et al. study. Retrieved 18 August 2026.

1. Robert M. Bond, Christopher J. Fariss, Jason J. Jones, Adam D. I. Kramer, Cameron Marlow, Jaime E. Settle and James H. Fowler. A 61-million-person experiment in social influence and political mobilization. Nature 489:295–298, 2012. doi:10.1038/nature11421. We could not access the full text. It is paywalled. All details about the study in this post come from the Smithsonian Magazine article, not from the paper itself.

Changelog

• 2026-08-18: First version drafted. Could not access the primary source.