The UK Just Banned Social Media for Under-16s. Australia Already Proved It Won't Work.

The UK is banning social media for under-16s. The ban’s own evidence suggests it won’t work. So why does it feel like progress?

📌 MARKETING / CONSUMER PSYCHOLOGY

On 15 June 2026, Keir Starmer announced what his government is calling an “Australia-plus” ban on social media for under-16s. Every major platform is in scope: TikTok, Instagram, Threads, Facebook, X, YouTube, Snapchat, Reddit. Gaming services and livestreaming platforms face additional restrictions. Disappearing messages and chats with adult strangers are targeted across all platforms. Personal messaging services like WhatsApp and Signal are not intended to be covered, though they fall within the government’s own definition. Romantic AI chatbots are banned for under-18s. General AI assistants like ChatGPT, Gemini, and Claude are not banned at all. For 16 and 17-year-olds, time limits are being considered. The government described social media as addictive, harmful to mental health, and a vector for bullying and dangerous content.

The consultation, titled Growing Up in the Online World, ran from March to 26 May 2026 and received 116,211 responses, making it the second-largest government consultation in history. The announcement came less than three weeks later. The government has promised a full analysis of those responses in the summer. The policy was announced before the analysis existed.

If you are a parent, this feels like someone is finally doing something. Nine out of ten parents who responded to the consultation expressed support for a minimum age of 16. But 72% of children who responded to the same consultation expressed concern about feeling socially excluded if restrictions were introduced. The government chose to emphasise the parents. If you are a behavioural scientist, this feels like someone is doing the wrong thing with great confidence. And if you are a marketer, this feels like watching a government discover, far too late, that the engagement architecture you have been optimising for the last fifteen years was always going to win.

This article is not a defence of social media. The platforms have made choices that prioritise engagement over wellbeing, and those choices have consequences. But the UK government’s response reveals something more interesting than the problem it is trying to solve. It reveals how adults misunderstand behavioural systems, how political incentives produce bad policy, and why banning a medium is structurally different from regulating a mechanism.

The ban is a policy admission that behavioural design won. It just doesn’t know what to do about it.

The first problem is definitional, and it is already leaking.

The government’s working definition of social media for the purposes of this ban is: “Any platform that allows for hosting user-generated content and permits public and private messaging.” The Online Safety Act 2023 already defines regulated services broadly as “user-to-user services” and “search services.” Read that carefully. It includes TikTok and Instagram, obviously. But it also includes WhatsApp, Telegram, and Signal. It includes Discord. It includes any gaming platform with a chat function. It includes platforms that most people do not think of as social media at all.

This covers websites and apps that enable users to communicate with others, upload content, or conduct searches. By that standard, the internet is social media. Email is social media. A Google Doc with comments enabled is social media.

The government has tried to narrow the scope by naming specific platforms. TikTok, Instagram, Threads, Facebook, X, YouTube, Snapchat, Reddit: the “Big Eight.” But naming platforms does not solve the definition problem, because the thing the government is actually worried about is not the platform. It is the behavioural pattern the platform enables. And behavioural patterns do not respect platform boundaries.

A fifteen-year-old banned from Instagram can achieve the same social comparison dynamics in a WhatsApp group chat. The variable reward schedule that makes TikTok compelling operates identically in Discord notifications. The fear of missing out that drives Snapchat use is present in any group communication channel. The architecture is the problem, not the address.

This is not a theoretical concern. The announcement specifically targets “disappearing messages” and “chats with adult strangers” as features to restrict across all platforms. That is an admission that the government knows the ban on named platforms is insufficient, and that the real regulatory target is a set of design patterns. But instead of regulating the patterns, they have chosen to ban the platforms. It is like banning specific brands of cigarettes while leaving nicotine unregulated.

The messaging app problem is the sharpest edge of this contradiction. WhatsApp and Telegram fall within the government’s definition. They host user-generated content. They permit public and private messaging. But they also use end-to-end encryption, which means the platform cannot see what is being shared. Enforcing a ban on under-16s using WhatsApp would require either undermining encryption (which the platforms have said would lead them to leave the UK market) or age-gating at the device level (which raises privacy concerns that make the Online Safety Act’s existing provisions look modest by comparison).

The government is either going to have to exempt messaging apps, which undermines the entire premise, or regulate encryption, which is a different policy with different consequences. Either way, the definition problem is not a technicality. It is a structural flaw in the policy’s foundation.

Then there is the AI chatbot question, which exposes the definition problem in a different direction. The government has banned Instagram but not ChatGPT. It has banned TikTok but not Character.ai. General-purpose AI assistants do not fit the definition of social media because they generate content rather than host user-generated content, and the “messaging” is between user and AI, not between users. So a fifteen-year-old can have a conversation with an AI chatbot, spend hours interacting with it, form emotional attachments to it, and receive a continuous stream of personalised, engaging content, and the ban does not apply.

The government has separately banned “romantic companion” AI chatbots for under-18s, targeting platforms designed to simulate sexual relationships or roleplay. That is a higher age threshold (18, not 16) than the social media ban, which is itself revealing: the government apparently considers a teenager talking to a romantic AI more dangerous than a teenager scrolling TikTok. But general AI chatbots are not banned. The government has said it is “considering broader age restrictions for AI chatbots” and is bringing them under the scope of the Online Safety Act. That is a holding position, not a final one.

The direction of travel is clear. If AI chatbots become a replacement social environment for teenagers, and they already are — children are using Character.ai and similar platforms as social spaces — the definition will expand. But the government has not acted yet, because the political urgency is about the platforms that parents recognise, not the ones that parents do not understand. The ban targets the social in social media while ignoring the media: the content, the engagement loops, the dopaminergic architecture. An AI chatbot can deliver all of that without any other users involved. A teenager having a deep conversation with an AI is arguably more isolating than a teenager scrolling Instagram, but the government has banned the one and permitted the other, because the policy is driven by parental anxiety, not by behavioural analysis.

II. Australia Tried This. It Didn’t Work.

The UK government has been explicit that Australia’s experience informed its decision. So it is worth examining what actually happened there.

Australia’s social media ban for under-16s went live on 10 December 2025. It applied to any service that allows social interaction between two or more users and permits users to post material. The scope was broad: TikTok, YouTube, Snapchat, X, Instagram, Facebook, and more. In the first days, over 4.7 million social media accounts were deactivated, removed, or restricted.

That sounds decisive. But the eSafety commissioner’s own compliance update, published in March 2026, told a different story. A “substantial proportion” of under-16s in Australia kept their accounts, managed to create new ones, or bypassed age-gating systems. The eSafety commissioner has since opened enforcement investigations into five major platforms: Facebook, Instagram, Snapchat, TikTok, and YouTube, with civil penalties of up to $49.5 million for non-compliance. The platforms are being punished for failing to enforce a policy that the policy’s own data shows cannot be enforced.

This is not a minor enforcement gap. It is a majority non-compliance rate. The policy failed to achieve its primary objective for the majority of its target population within three months of implementation.

Why? Because teenagers are not passive recipients of policy. They are active agents operating within a system, and when you erect a barrier, they find a way around it. This is not defiance. It is the predictable outcome of applying a prevention-focused intervention to a population that is developmentally oriented toward promotion.

Regulatory Focus Theory, developed by E. Tory Higgins, distinguishes between two motivational orientations. Promotion focus is concerned with growth, advancement, and the presence or absence of positive outcomes. Prevention focus is concerned with safety, responsibility, and the presence or absence of negative outcomes. Adolescents, whose prefrontal cortices are still developing, are disproportionately promotion-focused. They are wired to seek novelty, social reward, and autonomy. A ban is a prevention-focused intervention: it says “this is dangerous, avoid it.” But a promotion-focused brain hears “this is restricted, therefore it must be valuable.” The ban does not remove the motivation. It amplifies it.

This is not speculation. It is the observed outcome in Australia, and the UK government knows it. The consultation that produced this policy received over 100,000 responses, many of which would have cited the Australian data. The government processed those responses in less than three weeks. That is not evidence-based policy. That is a political timeline.

III. The Evidence Is Weaker Than You Think

The government’s rationale rests on a claim that social media is addictive, harms mental health, and exposes children to dangerous content. These claims are widely believed. They are also significantly more contested than the public conversation acknowledges.

The most rigorous meta-analyses of social media’s effect on adolescent mental health have consistently found that the effect size is small. Very small. Amy Orben and Andrew Przybylski’s 2019 study in Nature Human Behaviour, which analysed over 350,000 adolescents using specification curve analysis across three large-scale datasets, found that the association between digital technology use and well-being accounted for at most 0.4% of the variation. To put that in context, the negative effect of wearing glasses on adolescent wellbeing is larger. The negative effect of eating potatoes is comparable. Their 2022 follow-up in Nature Communications found that effects varied across developmental windows: girls showed negative links at 11-13, boys at 14-15, but at other ages the link was not statistically significant. The effect is real. It is also tiny, inconsistent, and bidirectional.

This does not mean social media has no effect. Small effects at population scale matter. But it does mean that the narrative of social media as a primary driver of a teen mental health crisis is not supported by the strongest available evidence. The crisis is real. The attribution is contested.

Jonathan Haidt’s The Anxious Generation has popularised the claim that smartphones and social media are the primary cause of rising adolescent anxiety and depression. But Haidt’s thesis has been substantially challenged by researchers who point out that the timing of the mental health decline does not align cleanly with social media adoption, that the decline predates smartphones in several countries, and that correlation between screen time and mental health outcomes does not establish causation. Przybylski has described the evidence base as “weak”, arguing that extraordinary claims require extraordinary evidence that is currently lacking. A large-scale analysis co-authored by Przybylski spanning 72 countries found no consistent global association between internet or social media adoption and changes in well-being. Candice Odgers, a psychology professor at UC Irvine, has argued that the idea of digital technologies “rewiring” children’s brains is not scientifically supported. Critics draw parallels between the current alarm and previous moral panics surrounding television and video games, where initial fears outpaced scientific evidence.

The government’s use of the word “addictive” is itself worth interrogating. Clinical addiction has specific diagnostic criteria: tolerance, withdrawal, continued use despite negative consequences, loss of control. Most adolescent social media use does not meet these criteria. What researchers observe is closer to compulsive use: habitual, sometimes excessive, but not pathological. The distinction matters because it changes the policy response. If social media is addictive, the correct response is restriction, as with substances. If it is compulsive, the correct response is design regulation, as with gambling.

The government has chosen the addiction frame because it is politically useful. It simplifies the narrative, assigns clear blame, and justifies decisive action. But it is not what the evidence supports, and building policy on a misdiagnosis produces interventions that target the wrong mechanism.

This is a textbook appraisal error. Appraisal Theory of Emotion, one of the foundational frameworks for understanding how humans generate emotional responses, distinguishes between primary appraisal (evaluating a stimulus as threatening, challenging, or irrelevant) and secondary appraisal (evaluating one’s capacity to cope). The government’s primary appraisal of social media is “threat.” Its secondary appraisal is “we cannot cope with this threat through existing means, therefore we must remove the stimulus.”

But a more accurate appraisal would be: social media is a challenge, not a threat. The design patterns that make it compelling are well understood. Variable reward schedules, social comparison triggers, infinite scroll, notification architectures: these are engineering choices, not natural phenomena. They can be regulated. The government has the tools to do this through the Online Safety Act and Ofcom’s existing powers. It has chosen not to use them, because a ban is simpler to announce and easier to understand.

The appraisal has been distorted by political incentive. And the policy follows the appraisal, not the evidence.

IV. The Architecture Is the Problem

If the ban is the wrong response, what is the right one?

The answer requires understanding what social media actually does to human behaviour, and the answer is not “it harms children.” The answer is that social media platforms have been engineered to exploit predictable features of human cognition, and those engineering choices affect adults and children differently because adults and children have different cognitive architecture.

This is where Dual Process Theory becomes essential. Daniel Kahneman’s framework distinguishes between System 1 thinking (fast, automatic, intuitive) and System 2 thinking (slow, deliberate, analytical). Social media is a System 1 environment. It is designed to trigger rapid emotional responses: outrage, amusement, envy, desire. The engagement loop is a System 1 loop. You see something, you feel something, you react. There is no space for System 2 evaluation.

Adults have more developed System 2 capacity. They can, with effort, pause, evaluate, and reappraise. They can recognise that the perfectly curated Instagram post is a performance, not a reality. They can notice that the outrage-bait tweet is designed to provoke them. They can, sometimes, put the phone down.

Adolescents cannot do this reliably, not because they are stupid, but because their prefrontal cortex is not fully developed. The prefrontal cortex, which governs executive function, impulse control, and deliberative reasoning, does not mature until the mid-twenties. This is not a controversial claim. It is basic developmental neuroscience. What it means is that adolescents are disproportionately reliant on System 1 processing in environments specifically engineered to exploit System 1 processing.

The government has identified this correctly. But its response is to remove the environment rather than change the engineering. And this is where the policy fails, because the environment is not the variable. The architecture is.

Consider what “regulating the architecture” would actually look like. It would mean banning infinite scroll for users under 18. It would mean requiring platforms to offer chronological feeds as the default, with algorithmic feeds as an opt-in. It would mean restricting variable reward schedules in notification systems. It would mean requiring platforms to disclose the engagement mechanics they use and submit to independent audits of their design choices. It would mean treating engagement architecture with the same regulatory seriousness as gambling mechanics, because the underlying psychology is the same.

None of this is in the government’s announcement. Instead, we get a ban on access, paired with vague commitments to restrict “high-risk features” at some undefined future date. The features are the problem. The ban is the distraction.

Social Identity Theory adds another layer. Adolescents are in the process of forming their identities, and social media provides a mirror for that process. The Socialiser, to use the STAR framework’s terminology, is driven by relatedness, and social media is the most powerful relatedness engine ever built. The Adventurer is driven by autonomy, and the ability to curate an online identity is a form of autonomy that adolescents crave. The Thinker is driven by competence, and social media provides constant feedback on social competence. The Realist is driven by security, and the social media environment can provide a sense of belonging that some adolescents do not find offline.

Banning social media for under-16s does not remove these developmental needs. It removes the most accessible outlet for them. The government’s paired investments in music, culture, sport, and youth services are an implicit acknowledgment of this. They are saying: “We are taking away the thing you use to feel connected, but here are some alternatives.” That is not a policy. It is a substitution, and it assumes that the alternatives are equivalent. They are not.

V. The Gambling Parallel

The closest regulatory precedent for what the government should be doing is not tobacco or alcohol. It is gambling.

The UK Gambling Commission regulates gambling mechanics, not gambling access. It does not ban adults from gambling. It requires operators to disclose odds, restricts certain design patterns (such as autoplay and near-miss mechanics), mandates responsible gambling tools, and prohibits marketing that targets vulnerable populations. It regulates the architecture, not the access.

Social media platforms use the same psychological principles as slot machines. Variable ratio reinforcement schedules, where rewards are delivered unpredictably, are the most effective way to sustain compulsive behaviour. This is not a metaphor. It is the same mechanism. The notification that sometimes contains something exciting and sometimes does not is a slot machine lever. The pull-to-refresh gesture is a physical instantiation of the same action.

The Gambling Commission’s approach works because it targets the mechanism, not the user. It accepts that people will gamble and focuses on making the gambling environment less exploitative. The government’s social media ban does the opposite. It accepts the exploitative environment and focuses on restricting who can enter it. The environment remains unchanged. The exploitation continues. The only difference is who is exposed to it.

Cognitive Bias Theory explains why the ban feels like progress even though it is not. The availability heuristic makes the harms of social media vivid and easy to recall: the stories of cyberbullying, the links to self-harm, the anxiety of comparison. These are real, and they are emotionally compelling. But the availability heuristic also makes us overweight vivid risks and underweight diffuse ones. The risk of a teenager being harmed by social media content is vivid. The risk of a teenager being isolated from their peer group by a ban is diffuse. The risk of a government learning that it can ban communication technologies when it finds them inconvenient is even more diffuse. The heuristic biases us toward supporting the ban, even when the evidence suggests it will not work.

VI. Ian Russell’s Verdict

Ian Russell is the father of Molly Russell, a fourteen-year-old who took her own life in 2017 after viewing harmful content on Instagram and Pinterest. Her death became a landmark case in the debate about social media safety. The coroner ruled that Molly died “while suffering from the negative effects of online content.” The Molly Rose Foundation, established by her family, has been one of the most prominent voices calling for platform accountability.

After the government’s announcement, Ian Russell described it as “disappointing and rushed”. He called the proposed blanket bans “deplorable” and warned they could be “gambling with young people’s lives.” Instead of banning access, Russell advocates ending engagement-based algorithms, infinite scroll, and disappearing messages for young people, as well as preventing strangers from contacting children online. He believes that robust online safety legislation, rather than bans, is the most powerful tool to protect young lives.

This is the most significant sentence in the entire debate. The person with the greatest moral authority on this issue, the person who lost his daughter to the mechanism the government claims to be addressing, has said the government is targeting the wrong thing. Not that the problem is overstated. Not that nothing should be done. That the intervention is misdirected.

The Molly Rose Foundation has been more explicit. Andy Burrows, the Foundation’s Chief Executive, said the ban would “quickly unravel” and is not supported by evidence, as children might circumvent age checks and move to less regulated spaces. In a joint statement signed by 42 child protection charities and online safety groups, including the Molly Rose Foundation, it was emphasised that blanket bans are a “blunt response” that do not address the shortcomings of tech companies. Ian Russell highlighted that bans risk unintended consequences and “let social media platforms off the hook” by weakening the requirement for them to offer safe experiences.

Out of control algorithms. Not out of control teenagers. Not out of control platforms. The algorithms. The architecture. The mechanism.

The government has had the Online Safety Act since 2023. It has had Ofcom’s regulatory powers. It has had the ability to regulate algorithmic recommendation systems, to mandate chronological feeds, to require transparency in engagement mechanics. It has not used these powers with any urgency. Meanwhile, new research from the Molly Rose Foundation found that 47% of girls aged 13-17, and a third of all teenagers in that age range, encountered high-risk content related to suicide and self-harm in a single week, despite existing safety measures. The Online Safety Act is already failing. The government’s response is not to enforce it more effectively, but to add a ban on top of it. Instead, it has announced a ban that its own evidence base suggests will fail, on a timeline that precludes serious analysis, and called it a victory for children’s safety.

Ian Russell’s disappointment is not about the principle of protecting children. It is about the method. And his preferred method, targeting the algorithms, is precisely what the government has chosen not to do.

VII. What the Ban Actually Reveals

The ban reveals three things about adult behaviour that are more interesting than the policy itself.

First, adults consistently misdiagnose systems as substances. The government treats social media like a harmful substance that can be kept away from children, as though it were alcohol or tobacco. But social media is not a substance. It is a system, and systems do not respect boundaries. The behavioural patterns that make social media compelling operate across all digital communication. Banning TikTok does not eliminate the variable reward schedule. It just moves it to a different platform. The ban is a category error, and the category error is driven by an adult need to simplify complex systems into controllable variables.

Second, political incentives produce prevention-focused responses to promotion-focused problems. A ban is easy to announce, easy to understand, and easy to defend. “We are keeping children safe” is a one-line policy. “We are regulating the engagement architecture of digital platforms to reduce the exploitation of System 1 processing in adolescent populations” is not. The political incentive is toward the simple, visible intervention, even when the complex, invisible intervention would be more effective. This is not unique to social media. It is a structural feature of democratic governance, and it means that policy will consistently favour access restriction over design regulation.

Third, the ban is an admission of defeat. The government has access to the Online Safety Act, to Ofcom’s regulatory powers, to the expertise of behavioural scientists, to the precedent of gambling regulation. It has chosen not to use any of these tools in a meaningful way. Instead, it has announced a ban that Australia has already shown will not work. This is not a government that has solved the problem. It is a government that has given up on solving the problem and opted for the appearance of solving it.

The teenagers will keep scrolling. They will find workarounds, as Australian teenagers did. They will use VPNs, create accounts with false ages, migrate to platforms that are not on the banned list. The 60% non-compliance rate in Australia is not a prediction. It is a precedent.

And when the ban fails, the government will face a choice: double down on enforcement, which will require increasingly invasive age verification measures, or acknowledge that the approach was wrong and regulate the architecture instead. The first path leads to a surveillance infrastructure that should concern anyone who values privacy. The second path leads to the policy that should have been announced today.

VIII. The Real Conversation

The UK government has done something that feels like action. It has announced a ban, cited grieving parents, and positioned itself as the protector of children. These are politically effective moves. They are not behaviourally effective moves.

The real conversation is not about whether children should use social media. It is about what happens when a government discovers that the engagement architecture it allowed to be built is more powerful than the regulatory tools it has available. The ban is not a solution. It is a symptom. It is the policy equivalent of a parent taking away a teenager’s phone and wondering why the teenager is still unhappy.

The behavioural design won. The question is whether adults are ready to have the conversation about why.

David Chadderton spent his twenties and thirties teaching people how to make life-or-death decisions at forty thousand feet. He now applies the same principles to consumer psychology, which, depending on the brief, can feel equally high-stakes. He’s the creator of the STAR Framework and the author of The STAR Framework: Rewriting the Rules of Consumer Engagement (NYC Big Book Award 2025), The STAR Operating System: Decode Mindset, Understand Motivation, Transform Human Behaviour, and Dear Algorithm, It’s Not Me, It’s You. By day, a Chief Marketing Officer. By night, a behavioural science obsessive who writes The Unoptimised Human because he can’t stop thinking about why people do what they do.

116,000 Responses. Zero Analysis. One Ban. What the UK’s Social Media Ban for Under-16s Reveals About Behavioural Design, Teenage Brains, and Policy Failure

I. What Is “Social Media” When You Try to Ban It?