Here is the asymmetry at the heart of children's relationship with digital systems: the internet is optimised, in real time, by some of the most sophisticated machine learning systems ever built. The human brain it is optimising against is, in children, still developing—in some critical capacities, not fully developed until the mid twenties.
When the U.S. Surgeon General issued an advisory on social media and youth mental health in 2023, one of its central findings was stark: children who spend more than 3 hours a day on social media face double the risk of poor mental health outcomes. Three hours. That is less than a single afternoon session on TikTok for many teenagers. And the systems are not designed with those teenagers in mind. They are designed with engagement in mind.
Children—whose prefrontal cortex is not yet capable of the same executive function, impulse regulation, and long term consequence reasoning as adults—are among the most effective targets of engagement optimisation systems ever deployed at scale. This is not a side effect. It is, from the platform's perspective, a feature.
What "Still Developing" Actually Means
The prefrontal cortex—responsible for impulse control, risk assessment, long term planning, and resisting immediate reward in favour of delayed benefit—continues developing into the mid twenties. Dopamine reward loops, which platforms exploit through variable reward schedules (the pull to refresh mechanism is designed to function like a slot machine), are more powerful in adolescent brains. The social comparison dynamics that Instagram amplifies trigger stronger responses in teenagers. The fear of missing out that platform notification systems deliberately engineer is more acute in identity forming years than in settled adulthood.
Internal research at Meta, revealed by whistleblower Frances Haugen in 2021, showed that Instagram was making teen girls' body image issues worse. The company did not act on this research because acting would have reduced engagement. That is not an anomaly. It is the logical output of a system whose primary objective is engagement, full stop.
Three Stages, Three Different Needs
Younger children need simple, visible rules: AI is a helper, not a person. Ask an adult before using a new tool. Don't share personal information. Adverts and recommendations are trying to influence. Co-use is protective—use devices alongside them, narrate what you're seeing, keep rules concrete.
Tweens are in identity forming years—highly susceptible to social comparison and beginning to encounter recommendation systems at their most influential. They need pattern recognition: why am I seeing this? Is this an ad? Is this trying to make me feel something? Ask questions before giving lectures. Curiosity is more effective than restriction at this stage.
Teenagers need system literacy and ethical reasoning: how AI can mislead, how feeds profile behaviour, how advertising and influence work, how to verify sources, how to use AI ethically for schoolwork. Move from control to coaching. You cannot control what a seventeen-year-old does online. You can equip them to navigate it with greater awareness.
The aim is not fear. The aim is agency. Fear produces avoidance—children protected from the system until the moment they are not. Agency produces children who understand the system well enough to make choices within it, for the rest of their lives.
