Framing
Not How Long, but What
The dominant public discussion of social-media harm to minors has centered on outright bans and time limits. Australia legislated a minimum age of sixteen for several platforms in 2024; other jurisdictions are considering similar measures. The interventions are largely cosmetic — minors route around age gates with VPNs, borrowed accounts, and minor data manipulation — and they target the wrong variable.
The harm to a young user is not measured primarily in minutes of exposure; it is measured in the quality of what is encountered during those minutes. Five minutes of grotesque content is not improved by being only five minutes. The indicative question for this problem space is not how do we shorten the session? It is how do we govern what enters the session at all?
Section 1
The Wrong Variable
Duration-based interventions — age gates, screen-time caps, scheduled lockouts — are appealing because they are easy to legislate and easy to demonstrate. A regulator can show a clock, a parent can show a meter, a platform can show a banner. None of these instruments, however, addresses the mechanism by which the harm is actually produced.
The mechanism is recommendation. Algorithmic feeds surface a sequence of items optimized for emotional response and continued engagement. The harmful pattern — escalating exposure to body-image content, self-harm imagery, political extremism, parasocial dependency — is produced not by the existence of the session but by the composition of what fills it. A child who spends twenty minutes inside a recommender system optimized for outrage is, on the evidence of every adolescent-mental-health study published since 2018, in a different category of risk than a child who spends two hours inside a recommender system optimized for craft instruction or family communication.
Treating duration as the variable to be regulated implicitly accepts that the recommender system itself is fixed. It is not fixed. It is a configurable surface, and the configuration is the thing that determines whether the time spent on the platform is harmful, neutral, or beneficial.
Section 2
The Relevant Indicative Fact
Platforms have demonstrated full capacity to alter what minors see when they are compelled to. The capability is not the constraint; the compulsion is.
Meta's "PG-13 inspired" content settings, rolled out to teen accounts in the United States, United Kingdom, Canada, and Australia in October 2025 and made the global default in April 2026, materially changed which content was recommended to under-eighteen users. The system has serious limitations — the Motion Picture Association forced Meta into a public disclaimer that the two ratings systems are unrelated — but it falsified the claim that this kind of intervention is technically infeasible. It is not infeasible. It is contingent on whether platforms are required to do it.
The technical infrastructure for cross-platform content rating exists in a more rigorous form than the consumer-facing announcements suggest. ROOST, a non-profit launched in February 2025 with Discord, OpenAI, Roblox, and Bluesky as founding partners, has released open-source moderation models and tooling explicitly designed for cross-platform use. The pieces required for a federated rating system — labelers, classifiers, filter-routing — are already shipped.
What does not yet exist is a regulatory mandate that platforms expose a standardized interface for third-party rating modules. That mandate is the design move that converts existing architecture into infrastructure.
Section 3
The Standard Objection
The skeptical case against content rating systems is that they have been tried and have failed. The World Wide Web Consortium's Platform for Internet Content Selection (PICS), released in 1996, was precisely the kind of open standard now being proposed again: anyone could publish labels, any browser could filter on them, and rating bodies could compete. It was largely abandoned, as was the Internet Content Rating Association that ran on top of it.
These failures, however, were failures of voluntary adoption. There was no requirement that publishers label their content, and no requirement that platforms honor labels published by third parties. The infrastructure was technically valid; the incentive structure made it dead on arrival. A system in which compliance is optional, in a market where attention is monetized by avoiding compliance, will not be adopted.
The remedy is not to abandon the technical pattern. It is to attach the mandate the prior generation of the pattern was missing. The lever required here is policy — not innovation, because the innovation already exists; not public communication, because the public is already aware of the harm.
Section 4
Three Candidate Interventions
Three candidate interventions, sketched for argument. Each is named by the lever it requires.
A federated rating layer
Open-source rating modules, certified for methodology by independent bodies, generate auditable labels on posts and direct messages. Users select which modules govern their feed; platforms above a defined scale are required by regulation to expose a standardized interface.
Algorithmic transparency
Platforms required to expose user-facing controls over the recommender system itself — what is amplified, what is suppressed, what the algorithm is optimizing for. The dial, not just the output. A condition of operating at scale, not a courtesy.
Parental content tethering
Legal parental authority over content filters for minors under eighteen, not merely over screen-time caps. Quality of exposure is the variable that determines harm; current parental controls almost exclusively address duration, which is the wrong primitive.
None of the three is sufficient on its own. The federated rating layer addresses the supply side — what gets labeled and made filterable. Algorithmic transparency addresses the configuration side — what the recommender is doing with the labeled content. Parental tethering addresses the authority side — who gets to set the rules for a minor's session. The three operate at different layers of the same problem, and the design space is materially better with all three than with any one of them.
It does not propose a particular rating taxonomy, a specific transparency requirement, or a particular parental-authority statute. The point is the structural argument: that the levers available are policy and innovation, not exhortation; that the wrong variable is duration, not exposure quality; and that the technical capacity exists and is waiting on a mandate.
Appendix A
References and Source Data
On the harm pattern
- Haidt, J. (2024). The Anxious Generation. Penguin Press. Synthesis of the adolescent-mental-health evidence and its alignment with smartphone-based social media diffusion.
- Twenge, J. M., et al. (2018). Increases in depressive symptoms, suicide-related outcomes, and suicide rates among U.S. adolescents after 2010 and links to increased new media screen time. Clinical Psychological Science, 6(1).
- Allcott, H., et al. (2022). Digital addiction. American Economic Review, 112(7).
On the regulatory model
- Information Commissioner's Office (UK). (2021–2025). Age Appropriate Design Code — implementation and audit reports.
- Australian Government. (2024). Online Safety Amendment (Social Media Minimum Age) Act.
- European Union. (2022). Digital Services Act — provisions on recommender system transparency.
On the technical infrastructure
- ROOST. (February 2025). Founding announcement and open-source releases. Discord, OpenAI, Roblox, Bluesky founding partners.
- Meta. (October 2025 / April 2026). PG-13-inspired content settings for teen accounts. Public announcements and MPA response.
- World Wide Web Consortium. (1996). Platform for Internet Content Selection (PICS). Specification and history.
Companion pieces
- EconFaithAI: From Imperative to Indicative — the method this piece applies.
- EconFaithAI: The Morality of Language Models — problem space ii.
- EconFaithAI: Workforce Displacement — problem space iii.
- EconFaithAI: What Shapes Children — the descriptive study that motivates this prescription.