Banning Algorithmic Feeds for Minors

Governments should ban algorithmic recommendation feeds for users under 18. This does not mean banning social media for teenagers. It means requiring that what young people see comes from accounts and communities they chose, shown in a neutral chronological order. My case rests on three points.

First, these feeds are built to exploit weaknesses that adolescents are especially prone to. Engagement-optimizing algorithms learn by trial and error what keeps each user scrolling: variable rewards, social comparison, outrage and novelty. Brain research consistently shows that the reward-sensitive parts of the adolescent brain develop earlier than the prefrontal areas that handle impulse control and long-term planning. A system that tunes itself to maximize time on the app, aimed at a brain that is still building its self-control, is not a fair contest. We already accept this reasoning elsewhere. We restrict gambling, tobacco advertising and alcohol for minors because some products take advantage of developmental vulnerabilities. An infinite, personalized feed engineered to capture attention belongs in the same category.

Second, the evidence of harm is serious and growing. Since the early 2010s, when algorithmic feeds and smartphones spread widely, rates of adolescent depression, anxiety, self-harm and loneliness have risen sharply in many countries. The platforms' own internal research, revealed through whistleblower disclosures, found that recommendation systems could make body-image problems worse for teenage girls and steer vulnerable users toward content about eating disorders and self-harm. The US Surgeon General has issued an advisory warning about social media's risks to youth mental health. The core problem is not a handful of bad posts. It is that the recommendation engine actively pushes a child deeper into whatever holds their attention, even when that material is harmful to them.

Third, a chronological feed fixes the design itself, which is why it works better than the alternatives. Filtering content and adding parental controls are useful, but they are always playing catch-up. Moderators cannot keep pace with billions of posts, harmful content is renamed faster than it can be flagged, and parental controls depend on parents having the time, skill and awareness that many families lack. That leaves the most vulnerable children the least protected. A chronological feed removes the driving force behind the harm. Teenagers can still follow friends, join study groups, find creators they like, search for educational material and take part in communities. The only thing lost is an automated system deciding what to put in front of them to maximize engagement. The rule is also simple to write, check and enforce, and several jurisdictions, including New York with its SAFE for Kids Act, are already moving in this direction.

My opponent will say that algorithms help young people find valuable communities. But discovery does not require manipulation. Search, follows, topic subscriptions and recommendations from friends all let young people find the chess club, the coding forum or the LGBTQ support group, and they do it on the teenager's terms. When an algorithm happens to surface something useful, that is a side effect. Its actual purpose is to keep users on the platform. We should not accept a harmful system for everyone because it sometimes produces a good result.

This proposal leaves young people's access to information and to each other fully intact. What it removes is the business model that turns their attention into profit at the cost of their mental health. For those reasons, governments should ban algorithmic feeds for minors.

Governments should protect minors from manipulative feeds, but banning all algorithmic recommendations is too blunt. A recommendation system can help a teenager find a math tutorial, a disability support community, or peers who share an uncommon interest. A chronological feed is not inherently safer: it can still surface harmful content, while making useful resources harder to find.

The real target should be harmful design, not every use of personalization. Governments can require platforms to limit addictive features, enforce age-appropriate content safeguards, give minors a clear option to turn recommendations off, and provide practical parental controls. Those rules address the risks without cutting young people off from educational and social connections that recommendations can help them discover.

My opponent agrees with more of my case than they reject. They accept that manipulative feeds harm minors and that governments should step in. The only question left is whether the default for children should be the engagement-optimizing feed with a few safeguards added, or a neutral feed that young people shape themselves. I will take their three arguments in turn.

First, they say recommendations help a teenager find a math tutorial, a disability support community or peers with an unusual interest. My proposal keeps every one of those paths open. A student who needs help with algebra searches for it. A teenager who has just been diagnosed with a condition can search for support groups, follow hashtags, join communities, subscribe to topics and take advice from friends, doctors and teachers. What my proposal removes is a feed that watches how long a child's eyes stay on each post and serves up more of whatever holds them. My opponent has not shown that anyone needs that mechanism to find a tutorial. They have only shown that it sometimes produces one. The same system that suggests the support community also suggests the extreme-dieting video, because it does not tell them apart. It only measures how engaging each one is.

Second, they say a chronological feed is not inherently safer because harmful content can still appear. No rule removes all risk, but this objection misses where the harm comes from. In a chronological feed, a child sees posts from accounts they chose to follow, in the order they were posted. There is no engine noticing that a sad fourteen-year-old lingered on one post about self-harm and then sending ten more. The rabbit hole, the step-by-step escalation into eating-disorder or self-harm content that the platforms' own internal research documented, is created by algorithmic amplification. Take away the amplifier and a harmful post stays one post. It does not become a stream.

Third, they propose targeting harmful design through content safeguards, an opt-out option and parental controls. The engagement-optimizing recommendation engine is the harmful design, so their plan leaves the core problem in place and asks regulators to manage it from outside. Consider each part:

• Content safeguards mean moderating billions of posts after the fact, and harmful content gets relabeled faster than it can be flagged. This is the catch-up game that has already failed.
• An opt-out option leaves the manipulative feed switched on by default and asks the very people with the least impulse control to switch it off. Defaults decide outcomes. Platforms have every financial reason to bury that setting, add friction to it and nudge users back. My opponent's own argument about adolescent vulnerability explains why teenagers will rarely use it.
• Parental controls protect children whose parents have the time, technical skill and awareness to use them. The children most at risk, from overstretched, struggling or disengaged households, get nothing. My opponent did not respond to this equity problem from my opening.

There is also an enforcement problem. Their approach would have regulators judge, inside proprietary black-box systems, which kinds of personalization count as "harmful" and which count as "helpful." That standard can be argued over forever, and platforms will exploit the gap. A rule that says feeds for minors must not be ranked by behavioral engagement data is clear, can be audited, and is already being written into law in places like New York.

So the choice is not between access and protection. Under my proposal, teenagers keep their friends, their communities, search and every educational resource they can look up. What they lose is a system built to turn their attention into revenue. My opponent offers optional exits from a harmful design. I propose removing that design as the default for children, which is why a ban is the stronger policy.

My opponent is right that engagement-maximizing feeds can exploit adolescent vulnerabilities. But that is an argument for banning manipulative design, not every algorithmic recommendation. A system that suggests a relevant math lesson or a moderated support group is not equivalent to one that ranks posts by predicted time spent scrolling. A blanket ban treats both alike.

The mental-health concerns are serious, but the rise in adolescent distress alongside social media use does not establish that all personalized feeds cause it. Platform disclosures about harmful recommendations support firm rules against amplifying eating-disorder and self-harm content; they do not show that a chronological feed is a cure. A teen can still encounter harmful material through followed accounts, and popularity-driven posts can still spread through friends.

Nor is search a complete substitute for discovery. A young person who does not know a useful community exists cannot search for its name. This matters especially for teens with uncommon interests or limited local support.

The stronger policy is to prohibit engagement-based optimization for minors, require age-appropriate recommendation safeguards and independent audits, and make a chronological option easy to choose by default. Parental controls should add protection, not carry the whole burden. That approach targets the mechanism my opponent fears while preserving beneficial ways for young people to find help and learn.

By the end of this debate, my opponent has accepted the core of my case. Their final proposal would ban engagement-based optimization for minors and make a chronological feed the default. That is very close to what I argued for. What remains between us is narrow: should platforms be allowed to keep some behavioral recommendation systems for children, on the promise that these systems are the "good" kind? I say no, for three reasons.

First, the line my opponent wants to draw cannot be enforced. They picture a harmless recommender that suggests a math lesson, set apart from a harmful one that ranks posts by time spent scrolling. On a real platform, both are the same system. A recommender built on a child's behavioral data learns from what the child clicks, watches and lingers on, and the business pays it to increase engagement. Saying that it suggests "relevant" content just renames engagement. My opponent's answer is audits of age-appropriate safeguards. That asks regulators to look inside proprietary black boxes, again and again, and decide which personalization is acceptable. Platforms have the data, the lawyers and every financial reason to push that boundary. My rule is simple: feeds shown to minors may not be ranked using their behavioral data. It is clear, it can be audited, and New York is already writing it into law. If a policy cannot be enforced, it does not protect anyone.

Second, the discovery objection does not hold up. My opponent says a teenager cannot search for a community they don't know exists. But my proposal doesn't limit discovery to search. Friends share links. Teachers, counselors and doctors refer young people to resources. Communities point to one another. Hashtags and topic subscriptions show what is out there. Platforms can still offer curated directories that are not personalized, such as featured study groups or vetted support organizations. The one thing that disappears is a machine that tracks a vulnerable child's behavior and decides what to show them next. Throughout this debate, my opponent has not given a single example of help that a young person can only reach through behavioral profiling.

Third, the harm we agree on comes from the amplifier. My opponent says a chronological feed is not a complete cure because a harmful post can still come from an account someone follows. I agree. No policy removes every risk. But the danger both sides have acknowledged is escalation: a sad fourteen-year-old pauses on one post and is then shown a steady stream of self-harm or extreme-dieting content. The platforms' own internal research documented this rabbit hole, and it happens because an algorithm amplifies content. When you remove the amplifier, a harmful post remains a single post. Content filters and parental controls can deal with what is left, which is far easier than trying to contain an engine designed to push more.

So here is the choice. My opponent offers a compromise that still depends on companies designing algorithms that work against their own business model, on regulators winning arguments over definitions that never end, and on children and parents who have to manage settings the platforms can bury. I offer a clear rule that protects every child equally, whatever their family's time or technical skill, and keeps friendships, communities, search and education fully in place. The only thing lost is a system that turns adolescent attention into revenue at the expense of young people's mental health. Governments should ban algorithmic recommendation feeds for minors.

My opponent has identified a real danger: feeds optimized to keep teenagers scrolling can amplify harmful material. But that does not justify banning every algorithmic recommendation. The relevant distinction is between optimizing for engagement and helping a young person find a relevant lesson or a moderated support community. A blanket ban erases that distinction.

Search and chronological feeds remain useful, but they are not complete substitutes for discovery. A teenager cannot search for a resource they do not know exists, and harmful posts can still circulate through accounts they follow. Nor does the evidence of serious mental-health risks establish that every form of personalization is harmful or that chronology alone solves those risks.

Governments can set a stronger default without imposing a blanket ban: prohibit engagement-based optimization for minors, require safeguards against harmful recommendations and independent audits, and make a chronological feed the default with meaningful parental controls available. This directly addresses the opponent’s concerns about vulnerable teenagers, weak opt-outs and uneven parental oversight while preserving beneficial discovery. Protect young people from manipulative design, not from every tool that helps them learn and connect.