Between the homework app and the midnight feed

A settlement worth billions. Two jury verdicts. Bills advancing in Congress. Personalization is at the center of it all. Proposals now before Congress would give minors more control over personalized recommendation systems and limit design features tied to compulsive use. California and New York already restrict personalized feeds to minors without parental consent, and Australia has banned minors under 16 from social media outright. 

Courts are weighing in too: in March, juries in two separate trials found online services liable for harms tied to platform design, and Meta recently settled a lawsuit brought by 47 states alleging it knowingly designed Facebook and Instagram's recommendation systems to be addictive to teens.

These approaches span a real range, from opt-out controls to outright bans. What they share is a critical question: is a given use of personalization designed to serve a young person, or to maximize the time that young person spends? A homework app that offers an easier explanation because a student struggled with fractions last week is personalization. So is a feed that has learned exactly what keeps a teenager scrolling at midnight. Both use data about a young person to decide what to show next. They differ in what data they draw on, and to what end. That distinction is the one policy has the hardest time drawing. 

Personalization, including the recommendation systems and features that decide what a young person sees, is neither inherently beneficial nor harmful. It often involves collecting and ingesting large amounts of personal data, including sensitive data, that can shape what young people see and how they interact with a service or feature, so it carries real risk to their privacy, safety and wellbeing. Researchers and child safety advocates have documented how it can contribute to "filter bubbles" that limit exposure to diverse ideas and distort users' view of reality, or surface increasingly extreme material through "rabbit holes," a risk heightened for young people still developing the capacity for critical evaluation. Those concerns are well founded, and they explain much of the current legislative attention. What they do not support is treating every use of personalization as the same thing. 

For that reason, personalization shouldn't always be governed with an on/off switch. It often requires a dial or a dashboard. Personalization spans dozens of uses, including information discovery, content, search, gaming, education, AI and advertising. Its effects depend on the service, the data used, and what it's trying to accomplish. Turning personalization off indiscriminately wouldn't just restrict harmful use; it would also take away the homework app's ability to help a student where they are. A more workable approach is to limit the data a service uses for a particular purpose, avoiding cross-platform tracking or overly detailed profile-building some systems rely on, while still using basic signals to adapt an interface or content for a young user. 

A recent report from The Future of Privacy Forum highlights this same distinction between functional and higher-impact personalization. Functional implementations use basic data, like device type, IP address, location, or age, to do things like localize the experience, adapt the interface, or apply age-appropriate safety settings. Higher-impact implementations collect and use far more data, like browsing history, watch time, or usage patterns, to shape content for engagement or monetization instead of function or safety. That difference in scope is why a risk-proportionate approach works better than a one-size-fits-all rule. The risks vary too widely by data practice and use case for one solution to fit every situation. 

Often, it's the young people with the fewest offline alternatives who rely most on personalization's benefits, like a teenager in an isolated community using a recommendation system to find people who share their interests, a young person navigating identity without offline support who finds community through content search, a student in an under-resourced school whose adaptive learning tool is their only individualized instruction, or a young person with a disability for whom adaptive design is what makes participation possible.

None of this argues for leaving personalization unregulated. Effective policy requires moving past whether personalization should exist in youth-facing services toward more precise questions: How is it implemented? What data does it draw on? What purposes does it serve? 

A risk-proportionate approach built on those answers can reduce harm without discarding what well-designed personalization offers young people. Drawing those distinctions is harder than flipping a switch, but it can help shape the digital experiences that support the real-world needs of youth.