Australia Banned Teens From Social Media. Did It Work?

Opening

Dear reader, on December 10, 2025, Australia enacted a full ban on social media accounts for anyone under 16. A month later, the government announced “We deleted about 4.7 million accounts—this is a success.” Yet a household survey conducted around the same time found that 60-70% of kids still had accounts.

Same policy, opposite numbers. Some call it a success; others say almost nothing changed. So which side is lying? Let me tell you the answer up front: both are true. They’re just measuring different things. And hidden inside that gap is a more uncomfortable fact: this ban has never actually been tested on the very kids it targets.

Same Policy, Opposite Numbers—Why?

The 4.7 million figure the Australian government cited is the number of accounts platforms found and removed. Meta said it deleted about 550,000 accounts within a single day of enforcement. From the platforms’ perspective, that’s a clear action and an eye-catching number.

The 60-70% figure, on the other hand, measures something else entirely. According to a March 2026 compliance review by Australia’s eSafety Commissioner, a parent survey found that 69.1% of kids still had Instagram accounts, 69.4% still had Snapchat, and 69.3% still had TikTok. Only about one in three parents said their child no longer had an account.

The two numbers aren’t in conflict. They’re just answering different questions. The 4.7 million measures “what did the platforms do?” The 70% measures “what actually happened to the kids?” Even after an account is deleted, a kid can just make a new one, lie about their age, or simply browse without logging in. While the count of deleted accounts keeps climbing, the number of kids actually accessing these platforms may barely move.

This is a textbook case of what I always watch out for: a ‘vanity metric’¹—a number that looks big and impressive but has no real connection to the outcome the policy was actually meant to change.

Why did the government lead with 4.7 million? Because it’s the cleanest number for giving the impression that “something was done.” Prime Minister Anthony Albanese even called it “a source of Australian pride.” The 70% figure, by contrast, exposes the policy’s gaps, so it rarely gets a spot on the same stage. Watching which numbers make it to the podium and which get left off tells you exactly what a policy is confident about—and what it wants to hide.

For context, this trend isn’t unique to Australia. France’s lower house passed a bill in January 2026 banning social media for anyone under 15, and Denmark is heading down a similar path. In the US, 8 states have introduced regulations targeting minors. Australia can fine violating platforms up to AU$49.5 million. The whole world, in other words, is moving quickly in the same direction. Even the UK Prime Minister, who initially opposed such regulation, has recently swung around to considering a ban—that’s how lopsided the mood has become.

A Policy That Was Never Actually Tested

So does this ban actually improve kids’ mental health? Remarkably, we don’t know.

A paper published in May 2026 by Monika Neff Lind’s team at the University of California, Irvine, addressed this head-on. The team compiled the randomized controlled trials (RCTs)²** that had tested whether quitting or cutting back on social media improves mental health.** They pulled 36 studies from two existing meta-analyses and added 4 more found in an additional search in December 2025, for a total of 40.

Three facts stand out here.

First, not a single one of the 40 studies had an average participant age under 18. There were no participants under 16 at all—the youngest participant was 16. In other words, no experiment has ever been run on the exact kids the ban targets. French President Macron called this “what scientists recommend,” but the science that actually tested that population simply doesn’t exist.

For context, out of 973 studies the team screened, only these 40 met the inclusion criteria. Of those, only 8 studies had even a single participant under 18—and even those were recruited from university subject pools, meaning they were effectively college students. Data on the middle- and high-schoolers we actually want to understand simply doesn’t exist anywhere.

Second, the trial periods were far too short. Restriction periods ranged from a single day to 3 months, but half lasted a week or less, and the average was 16.3 days. On top of that, 19 of the 40 studies didn’t involve quitting entirely—just partially cutting back usage time. It’s hard to extrapolate the effects of a years-long, blanket ban from experiments this brief.

Third, the effect itself was weak. Across the two meta-analyses, the average effect was statistically indistinguishable from zero, or if it existed, it was tiny (effect size³** g≈0.17).** Of the 40 studies, 8 found no effect, and 8 actually found that well-being got worse. Combined, that’s 40% showing “no effect or a backfire.”

Here’s what’s even more interesting: in most of these studies, participants knew they were “taking part in an experiment about quitting social media.” With the expectation that “quitting will make things better” already built in, the design was tilted toward positive results⁴. And yet the effect was still this small. Both meta-analyses also reported that the older the participants were, the larger the effect of restriction. That’s the opposite of what you’d expect if teens were supposed to benefit the most.

In fact, even the conventional wisdom that “social media harms teen mental health” isn’t scientifically settled. The US National Academies concluded in a 2024 report that “the associations are small and inconsistent, and reality is more complicated.” In longitudinal studies, mood or prior symptoms sometimes predict later social media use—meaning the direction of causation can run opposite to what conventional wisdom assumes.

There’s also a measurement trap at play here. Many existing studies asked about both “social media use” and “depression” in the very same survey. Even when a correlation shows up, it’s hard to tell whether that’s a real relationship or just an artifact of how people tend to answer surveys. And if you leave out a third factor that affects both—like income—you end up drawing the wrong causal picture entirely. This is exactly why researchers insist we need experiments, not just correlations.

To borrow the research team’s phrase, we’re essentially administering an untested treatment to every child in the country at once. If this were a drug, it would be like skipping clinical trials and distributing it nationwide anyway—even though it might work, might do nothing, or might actually cause harm.

Of course, there’s evidence pointing the other way too. Davis and Goldfield’s 2025 study found that among 17-25 year-olds experiencing emotional distress, cutting back on use reduced depression, anxiety, and FOMO. So it can help young people under certain conditions. But that’s a very different story from a blanket ban imposed on an entire population.

Take It Away, and the Balloon Bulges Somewhere Else

What’s even more worrying than the lack of validation is the unintended side effects.

Australia’s law blocks account ownership, but it can’t stop kids from browsing content without logging in. So they keep their access to content while losing exactly the protections that came bundled with an account—content filters, parental controls, and the like. And if a kid creates a fake adult account by lying about their age, the algorithm starts treating them as an adult.

Age-verification technology poses its own problem. Australia lets users choose between a government ID, a selfie photo or video, or a bank login. Even in the Australian government’s own testing, automated age-estimation technology showed higher error rates for younger-looking faces and people of color. So the very technology meant to protect kids is structurally more likely to get it wrong for certain groups. Anyone without ID, or whose ID no longer matches their current appearance, bears an even heavier burden. The paradox is that the most vulnerable kids end up being misidentified the most.

On top of this, there’s the developmental nature of adolescence. Teenagers are especially sensitive to autonomy and to feeling respected. Rules imposed unilaterally from above often provoke backlash, increase rule-breaking and concealment, and erode trust with adults. This means a ban could increase conflict rather than reduce it.

Furthermore, many schools and youth organizations use social media for announcements and event information. Pushing kids off these platforms creates the contradiction of also cutting them off from information they actually need.

There’s yet another side effect. Kids pushed off major platforms may migrate to more loosely moderated spaces. Instead of disappearing from visible, well-lit places, they scatter into corners with weaker controls and less protection. Lawsuits arguing that the ban conflicts with free expression or children’s rights are already looming in multiple places. A single policy risks triggering enormous administrative costs and legal disputes.

People often compare this to a balloon. Squeeze one side and the air doesn’t vanish—it bulges out somewhere else. Squeeze accounts, and anonymous access and fake accounts balloon in their place.

Oz’s Lens

There’s a failure pattern I’ve seen more than any other while building go-to-market strategies: shipping a solution before the problem is properly defined, then announcing the mere fact of having shipped as if it were an outcome. The 4.7 million figure is exactly that. It’s impressive, but it still has zero connection to the outcome the policy promised—kids’ mental health.

There’s a question I always ask when working with data: “What is this number actually measuring?” The 4.7 million measures “what the platform did”; the 70% measures “what’s true for the kids.” When announcing results, people tend to pick the bigger number. So I always look for the metric that wasn’t announced before I trust the one that was.

Interestingly, even the paper’s authors were candid about their own conflicts of interest. One holds equity in a digital health company; others have ties like advisory roles at Headspace or YouTube. This kind of honest disclosure actually strengthens the credibility of their argument—it’s the posture of exposing your limitations upfront, rather than deciding on a conclusion first and fitting the evidence around it.

This is also the very first thing I get agreement on when defining success metrics in consulting: “What would we need to see to admit we’ve failed?” A success announcement with no failure criteria attached is, in practice, an announcement that promised nothing at all.

Taking technology away isn’t always the wrong call. But before taking it away, you have to define what you’ll measure, and how, to judge success. Without that, we end up testing an unvalidated prescription on children first. Making something better is harder than simply taking it away, but that’s usually where the real solution lies.

Closing

Let me sum up today’s issue in three lines: First, Australia’s “4.7 million” and “70%” are both true—they’re just measuring different things. Second, the teen ban has never actually been validated on kids under 16, the very population it targets. Third, taking something away often just inflates the other side of the balloon.

Next time you see a “success” announcement about a policy or a product, I’d suggest asking one question first: “What exactly is this number measuring?” That single question tends to shine a fairly precise light on the gap between the announcement and reality. If you’re curious to dig deeper into the ban’s effects, I’d recommend starting with Section 4 of the paper below (the part proposing an evaluation methodology).

💬 Have you ever seen a metric announced as a “success” that turned out to be way off from reality, at work or in everyday life? Tell me what the number was in the comments.

References & Further Reading

Primary sources

• Neff Lind, M., Schueller, S. M., & Odgers, C. L., “We don’t know how social media bans will affect youth but we’re doing it anyway!”, Frontiers in Developmental Psychology, 2026. : This is the backbone of today’s issue. Table 1, which summarizes the 40 studies, and Section 4, which proposes an evaluation methodology, are especially essential.
• Burnell, K. et al., “The effects of social media restriction: meta-analytic evidence from randomized controlled trials”, SSM-Mental Health, 2025. : Shows in numbers why the effect of restriction is so small.
• Ferguson, C. J., “Do social media experiments prove a link with mental health”, Psychology of Popular Media, 2024. : The source of the conclusion that the average effect is indistinguishable from zero.
• Davis, C. G., & Goldfield, G. S., “Limiting social media use decreases depression, anxiety, and fear of missing out in youth with emotional distress”, Psychology of Popular Media, 2025. : A study finding the opposite—that it helped certain young people. Worth reading alongside the others for balance.

Background

• eSafety Commissioner, “Social Media Minimum Age: March 2026 Compliance Update”, Australian Government, 2026. : The source of the ~70% retention figures.
• National Academies of Sciences, Engineering, and Medicine, “Social Media and Adolescent Health”, 2024. : The US National Academies’ conclusion that “the relationship is small and complicated.” Worth reading as the starting point of this debate.
• Jonathan Haidt, “The Anxious Generation”, 2024. : The flagship argument for the pro-ban side. Knowing the opposing view is essential for balance. Published in Korean translation as Bulan Sedae (The Anxious Generation).

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis — business data management and business analytics — while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master’s from Korea University’s Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

1. Vanity metric: A metric that looks big and impressive but has little real connection to the goal you’re actually trying to achieve. A classic example is high view counts that never translate into purchases. ↩

2. Randomized controlled trial (RCT): A method that randomly splits participants into two groups (treatment/control) to compare outcomes. Considered the strongest tool for establishing cause and effect. ↩

3. Effect size (g): A value indicating how large an effect is. Generally, 0.2 is considered small and 0.5 medium. At 0.17, this falls on the small side. ↩

4. Demand characteristics: A phenomenon where participants sense the intent of an experiment and change their behavior to match expectations, which can inflate results. ↩