The Problem With Banning Social Media Is That It's Social

New evidence suggests teenagers will not log off until their friends do too

In December 2025, Australia became the first country to ban under‑16s from holding accounts on Facebook, Instagram, Snapchat, TikTok, and seven other major platforms.¹ Four months in, a research team including Leonardo Bursztyn, Angela Duckworth, and Cass Sunstein went to find out whether it was working. Their working paper, “Why Bans Fail: Tipping Points and Australia’s Social Media Ban,” gives us the first structured look at what has happened and what it says is concerning. About a quarter of 14‑ and 15‑year‑olds are complying. The other three quarters are still posting, scrolling, and lying about their birthdates.

---

I supported the Australian social media age ban when it was announced, and I have been, for some years, an enthusiastic advocate of getting phones out of classrooms. Neither view has changed. But I also try to be the sort of reader who updates their priors when new evidence comes to light, and a recent study has nudged me to do precisely that. Not by overturning the case for a ban; the case still holds. What it has done, rather, is question something which is far more concerning for me: that the scourge of social media is not a problem a blanket ban can solve on its own, and that a policy which addresses the apps while ignoring everything else children are growing up inside risks delivering the appearance of protection rather than the substance of it.

The prohibition paradox: when social bans make behaviour more social

The interesting finding with this new study is not that the ban might be failing. The interesting finding for me is why it might be failing. It seems the law has not come close to reaching the tipping point at which enough teens change their behaviour that others are willing to follow. Put simply, it has declared a rule without shifting the social equilibrium.

Something similar happened a century ago with American Prohibition. The Eighteenth Amendment banned the manufacture, sale and transport of alcohol from 1920, but drinking was so woven into social life that the ban mostly pushed it sideways rather than stamping it out. Legal saloons closed and were replaced by speakeasies, illegal bars that became the worst‑kept secret in every city, with some estimates suggesting there were more speakeasies than there had been licensed bars before the ban.

The American Issue, Illinois Edition, Vol. XVII, April 1923. Anti-Saloon League of Illinois, Westerville, Ohio.

For many Americans, going out for a drink did not stop being a normal social pastime; it simply became a slightly more glamorous, rebellious one, backed by bootleggers and organised crime instead of breweries and bartenders. Prohibition shifted the law on paper but not the descriptive norm: people still expected their friends, neighbours and even politicians to drink, so the noble experiment collapsed under the weight of everyday non‑compliance.

Injunctive norms versus lived reality

In this new study, teenagers were asked, in plain terms, what share of their peers would have to leave social media before they would too. The mean answer was 69 percent. The current compliance rate is 27 percent. That gap is not a measurement artefact. It is the gap between an injunctive norm (what the law says you ought to do) and a descriptive norm (what you can see your peers actually doing). When the two point in different directions, the descriptive norm usually wins.

Cristina Bicchieri has been telling us this for twenty years: social norms bind when people both expect others to comply (empirical expectations) and expect that others think one ought to comply (normative expectations), often backed by social sanctions. Seen this way, the law is almost a textbook exercise in moving only one piece of the norm system. Parliament has loudly broadcast the injunctive norm (“under‑16s should not be on social media”) but teens’ empirical expectations remain that most of their friends are still on Snapchat, TikTok, or YouTube, and they do not expect disapproval from peers if they keep using them.² Their beliefs about what their friends are doing, and what their friends think is acceptable, have barely budged; the formal rule is just another piece of adult noise in the background.

The deeper problem is one of social composition. The researchers asked teens whether kids who comply with the ban are more or less popular than kids who do not. Forty‑seven percent said less popular. Five percent said more popular. The compliers are sitting in the social periphery; the non‑compliers are at the centre. Bicchieri and others have argued that norms are sustained not just by rules but by the payoffs attached to compliance and violation. Here, the payoffs are upside‑down: following the rule makes you less central to the social world that matters to you.

---

What worked with smoking, and what is happening here

The Australian ban seems to be the inverse of what brought adolescent smoking down, but not for the reason the popular-kids-quit-first folk theory suggests. There’s a strong argument to say that smoking declined because the meaning of smoking was changed and the rate of initiation was cut as a result. The 1998 Master Settlement Agreement restricted tobacco advertising; the CDC’s Truth campaign recoded smoking from cool to industry manipulation; states raised tobacco taxes and banned indoor smoking; graphic warnings appeared on packs; adult smoking declined in parallel and reduced household exposure. The network effect followed rather than led: as Haas and Schaefer’s longitudinal data on US adolescents show, peers exert their strongest pull on smoking initiation, not on cessation.

Adolescents rarely start without peer influence, but they will quit while their friends carry on (see Haas & Schaefer). That is the leverage point smoking policy hit. Once initiation falls, each cohort sees fewer smokers, and the lower rate sustains itself.

There is a separate evidence base for targeted network interventions, which is what the original draft was reaching for. Paluck, Shepherd and Aronow’s anti-conflict experiment across 56 New Jersey middle schools found that interventions with the most socially connected students ( “social referents,” identified by peer nomination) reduced disciplinary reports for student conflict by roughly thirty percent over a year, while equivalent interventions with randomly selected students did nothing.

The mechanism, on closer reading however, is not high status as such. Damon Centola’s network-tipping work argues that what matters is the density of redundant ties among committed minorities; status helps mainly because it correlates with connectedness, but the active ingredient is mutual reinforcement among early compliers.

It appears Australia has not laid the ground for this. There has been no advertising restriction analogous to the Master Settlement Agreement, no counter-marketing campaign that has done to TikTok or Instagram what Truth did to Marlboro, no taxation of attention, no parallel adult withdrawal. The ban is a single crude instrument, aimed at use rather than at meaning, and the descriptive norm is at near-saturation: 91 percent of unbanned 16–18-year-olds in Australia, 95 percent of US teenagers, on the same data. There is no pool of never-users for a meaning-shift to protect, and no structural counter-marketing recoding what the platforms signal. The mandate has, if anything, conferred a faint countercultural sheen on continued use.

And if the loudest voices speaking out against social media are middle aged men (usually on social media unironically) then the campaign probably has the wrong vector, and teenagers, who are excellent readers of social vectors, will not miss it. Adolescents have a fine ear for the genre; when the moral entrepreneurs are the in-group, they hear that this is a thing the dads want, which is reason enough to keep doing it.

This matters beyond one country’s drafting. It matters for any rule that depends on coordinated abstention or coordinated adoption: phone-free classrooms, homework platforms, dress codes, the language norms of a corridor. A mandate that ignores the network composition of its early compliers, and the cultural meaning of the practice it targets, fails in the same recognisable way. The thresholds are too high, the rule becomes the badge of the dutiful and the uncool, and non-compliance accrues prestige it would not otherwise have. The smoking case is instructive precisely because it shows what the work actually consists of; meaning-shift plus structural levers acting on initiation, with network sustenance following.

Two design choices in particular are a problem

The authors in this paper highlight two major problems. First, the ban operates on age, not grade. Adolescent peer groups are organised by classroom, not by birthday. A 14‑year‑old sitting in a class with several 16‑year‑olds is staring directly at peers for whom the law does not apply. Within‑classroom exposure to unbanned older peers jumps sharply between 14‑ and 15‑year‑olds, and so does the rate of use. Grade‑based cut‑offs would have given the law a fighting chance of changing what teens actually saw around them.

Second, enforcement falls on platforms, not on users and their guardians. Civil penalties of up to A$49.5 million attach to companies who fail to verify ages. (This is laid out in early coverage such this BBC explainer.) Teens and their parents face nothing. The teens have noticed: only 22 percent believe there are personal consequences for them. Non‑compliance carries neither an individual cost nor a visible social signal, which is the regulatory analogue of asking the child not to eat the marshmallow while leaving the marshmallow on the table. Qualitative work on norms suggests that when individual sanctions (social or legal) are absent, and everyone can see that violators are not punished, injunctive norms quickly hollow out.

The numbers are striking. Among the paper's 14-year-olds, fewer than one in ten classmates on average is aged sixteen or older; by 15, the figure is roughly four in ten; by 16, eight; by 17, nine. The same transition between 14 and 15 that sees a teenager move from a class of largely compliant peers to one of largely non-compliant ones also sees use inside the banned group jump from 57 to 68 per cent.

In fairness, the authors are careful not to read this as causal, and rightly so, but the coincidence is hard to ignore: the moment the typical banned teenager begins to share a classroom with a critical mass of unbanned older ones is also the moment compliance begins to crumble. Drawing the line by year of schooling rather than year of birth would not have made the ban watertight, but it would at least have stopped the law from running headlong into the social geography of secondary school.

There is also a more basic design flaw. The duty to police ages falls on a subset of large platforms, leaving adolescents free to migrate to services outside the law’s scope. Early reporting suggests many Australian teens have shifted time into group chats, gaming platforms, and services such as Discord, which they do not perceive as “social media” in the same way. (See “Kids locked out of social media apps after Australia’s world‑first ban.”) From the teenager’s point of view, their social graph is still intact; only the logos on the icons have changed. In normative terms, the descriptive norm (“everyone I know is still online with me”) remains the same.

All of this would be worrying enough if the ban were a well‑targeted instrument for improving wellbeing. The evidence there is more complicated than the policy debate admits. Large longitudinal studies of Australian adolescents find a non‑linear relationship between social media use and wellbeing: very heavy use is associated with poorer outcomes, but moderate use often coincides with better wellbeing than either abstinence or extreme use. A synthesis of data on Australian teens concludes that “one‑size‑fits‑all rules and blanket bans are unlikely to be effective,” with moderate engagement (more than none, much less than all day) linked to the most favourable wellbeing profiles. Even if the law had changed behaviour, its mental‑health returns would probably have been uneven.

Designing a ban that could actually work

So what would actually shift behaviour? The literature is reasonably clear, even if politically uncomfortable. You have to engineer the descriptive norm, not just declare the injunctive one. That means:

• Working on what compliance signals about you, so that following the rule does not push you to the social periphery.

• Recruiting high‑status early movers, as network‑based smoking interventions try to do with central students.

• Providing alternative social infrastructure (sports, clubs, online and offline spaces) that fills the time and connection the platforms used to fill, especially for those most reliant on digital communities.

• Pairing all of this with friction at the device level (defaults, screen‑time controls, age‑gating at operating‑system level) rather than only at the corporate level.

None of it is impossible. Most of it is standard public health and standard social‑norms engineering but very hard to get right. As a commentary in Nature puts it, Australia has created a “natural experiment” in how far you can get by legislating a social norm without reshaping the social networks that sustain it.

If your intervention does not change what teens see their peers doing, do not expect it to change what teens do. Prohibition lasted thirteen years before that lesson was admitted. We can hope this one moves faster.

Protecting my kids, without fooling myself

I have skin in this game. I would love my three children (all still under ten), to grow up without the worst of what social media has done to this generation of teenagers; the relentless comparison, the always‑on group chat, the sense that your social life can fall apart because of something you missed while you were asleep. I don’t want them marinating in that from the moment they leave primary school.

But wanting that world does not mean any ban that gestures in its direction is going to help. If anything, blanket bans shows how easy it is to get the symbolism of protection without changing the social reality children grow up inside. You can pass a law, announce a new age limit, even wipe a few million accounts, and still leave every ten‑year‑old looking up at a social universe in which “everyone who matters” is online together, just out of their legal reach.

So I support a ban. But the question is which one? A ban that raises the formal age, leaves the platforms intact, and leaves every fourteen‑year‑old in my children’s future cohort already practised at lying about their birthday is not really a ban; it is stagecraft. Worse, it risks pushing children into lying more convincingly about their birthdays, into the unlit corners of the internet, the encrypted group chats, the Discord servers no parent has heard of, the Telegram channels that have already absorbed much of what mainstream platforms nominally moderate away, while convincing the rest of us that the problem has been dealt with.

So I find myself in an awkward middle: I do want a ban, but I want a ban my own children won’t experience as a dare, or a joke, or a lesson in how early you have to start lying to be part of the group. I hope we can do something slower and less photogenic that actually shifts their social world than something headline‑friendly that mostly shifts our self‑image as responsible adults. The test of a policy like this is not whether it lets parents feel that they have drawn a firm moral line. It is whether, ten years from now, a child like mine looks around their classroom and sees a different everyday reality than the one teenagers are living in now.

An Anti-Saloon League of America poster, circa 1917.

1. For a clear overview, see reporting in the New York Times.

2. The BBC’s “one month on” piece captures this ambivalence well: “I feel free”: Australia’s social media ban, one month on.