When something goes wrong for a child online—when they encounter harmful content, spend hours in a distressing spiral, develop an unhealthy obsession with body image or extremist ideas—the conversation almost always focuses on what they saw. The post. The image. The video. The account. And the advice that follows is almost always about content: report it, block it, restrict it, filter it.
This is not bad advice. Content matters. But it addresses a symptom while leaving the cause entirely intact. The deeper hazard isn't what your child saw. It's why they kept seeing it.
A harmful post is content. A harmful loop is system design. These are different problems requiring different responses. Mixing them up means consistently giving parents advice that cannot solve the problem they're trying to solve.
Three Levels of Risk
Social media risk operates on three distinct layers. At the first is content: the posts, videos, images, and misinformation a child might encounter. The parent question: what is my child seeing? Appropriate responses: filters, reporting, source checks, conversation. At the second are algorithms and AI: ranking, recommendation, personalisation, escalation, and attention optimisation. The parent question: why do they keep seeing more? This is a structurally different problem—you cannot solve it with content advice because the problem isn't a piece of content, it's the system generating a continuous supply of it. At the third is private messaging: DMs, group chats, and direct contact channels where grooming, bullying, and coercion occur. Parent question: who can reach them?
The Los Angeles Case
In March 2026, a Los Angeles jury found Meta and YouTube liable in a youth social media addiction case—one of the first of its kind to reach a verdict. The jury awarded $6 million. The case was notable not for the amount but for what it focused on. The allegations weren't primarily about specific pieces of harmful content. They were about addictive design: recommendation systems, autoplay, infinite scroll—the features that create and sustain the loop rather than any individual post within it.
The legal question is shifting from "what content was posted?" to "how was the platform designed?"—and that shift should also change how parents and schools think about the problem. If the hazard is the design, the solution cannot be content moderation alone.
We have spent a decade giving parents content advice for an algorithmic problem. Teaching children to spot bad content while leaving the recommendation system intact is like teaching someone to dodge individual cars while they're standing in the middle of a motorway. Necessary. But not sufficient.
Addressing the algorithmic layer requires asking different questions: why did this appear? What signal did the system read? Resetting the feed—clearing watch history, using "not interested," deliberately following different types of content—gradually retrains the recommendation system. Understanding that you are dealing with an optimisation machine, not a neutral curator, is the precondition for navigating it with any real agency.
