Do the proposed youth-safety measures produce measurable reductions in harm once evasion, platform switching and privacy costs are included?

I support adopting enforceable, age-specific teen time limits and behavioral-advertising bans, with mandatory independent evaluation and revision if harms outweigh benefits. This is a normative judgment under uncertainty—not a claim that existing laws have already demonstrated lasting, cross-platform improvements. The relevant outcome is a young person’s total harmful exposure and well-being, not one platform’s declining usage.

First, experimental evidence establishes that reducing overall recreational screen use can improve psychological outcomes. A 2024 JAMA Network Open randomized trial involving 89 Danish families and 181 children and adolescents tested a two-week, family-wide reduction. The intervention improved the Strengths and Difficulties Questionnaire total difficulties score by 1.67 points relative to controls (95% CI: 0.67–2.68 points; Cohen’s d=0.53). However, participants were predominantly younger children, follow-up was short, and the intervention was broader than platform-enforced teen limits. This supports causal plausibility, not a proven statutory effect or an optimal daily cap. (pubmed.ncbi.nlm.nih.gov)

Second, advertising restrictions address monetization rather than merely relocating minutes. Behavioral advertising means targeting advertisements using observed or inferred personal behavior. The EU Digital Services Act supplies a concrete precedent: Article 28 prohibits profiling-based advertising when covered platforms know with reasonable certainty that a user is a minor; Article 19 generally exempts micro and small enterprises unless designated very large platforms. This is not a universal advertising ban. (eur-lex.europa.eu) The DSA became generally applicable on February 17, 2024. (commission.europa.eu) My economic inference is that restricting this revenue channel can weaken incentives to capture teens’ attention, although contextual advertising and other monetization preserve some incentives; well-being gains must therefore be measured, not assumed.

The strongest objection is displacement. A survey of 430 previously heavy adolescent gamers following China’s 2021 restrictions found that 59% shifted toward short videos, 14% used relatives’ identities, and 3% rented accounts. These self-reports demonstrate substantial substitution, but do not establish whether total harmful exposure rose, fell, or remained unchanged. (pubmed.ncbi.nlm.nih.gov) Cross-platform displacement—activity moving between services or accounts—erases benefits only if replacement harms offset the harms prevented; its existence alone cannot answer that question.

I propose separate enforcement and evaluation for ages 13–15 and 16–17, preserving essential educational and support access. Age assurance means estimating or verifying age or age range: require minimal-disclosure credentials, retention limits, demographic accuracy audits, and appeals, consistent with EDPB principles. Privacy safeguards require verification, not vendor assurances. (edpb.europa.eu)

The decisive test should be an independently run longitudinal study—following the same participants over time—with consented, privacy-protected cross-platform measurement of harmful exposure, repeat harassment, compulsive use, and well-being. Track alternative accounts, age misreporting, less-moderated services, and AI companions. Randomized or staggered implementation should separately estimate time limits, advertising bans, recommender-system changes—the systems selecting and ranking content—and contact/reporting protections. Count privacy loss and unequal exclusion alongside benefits. Adoption should initiate accountable harm reduction, not declare success from lower usage.

B’s strongest point is that reducing recorded activity does not establish a net safety benefit, and evidence from voluntary interventions cannot directly identify the effects of compulsory teen limits. Indeed, Wang’s June 2026 study deserves a sharper qualification: its abstract reports suggestive indications of reduced mental well-being, not merely uncertain improvement. That is a genuine warning against equating less Internet use with better health. (sciencedirect.com)

Nevertheless, the same nationally representative difference-in-differences study estimates that China’s restrictions reduced total Internet time by 22% on the intensive margin despite incomplete compliance. This supports the narrower proposition that displacement need not replace every prevented minute; it does not establish reduced harmful exposure. My position remains a normative recommendation for enforceable, proportionate limits with mandatory evaluation—not an endorsement of China’s severe gaming schedule or a claim that its health outcomes vindicate restrictions. (sciencedirect.com)

Additional experimental evidence addresses B’s concern about relying exclusively on a family-wide device-surrender experiment. Davis and Goldfield’s randomized trial reported 220 participants aged 17–25 with emotional distress; the intervention requested smartphone social-media use of one hour daily for three weeks. Smartphone tracking measured use, and participants reported greater improvements in depression, anxiety, fear of missing out and sleep than controls. This is closer to the proposed activity restriction, but remains voluntary, short-term, largely outside the younger-teen population, and dependent on subjective health outcomes. It neither proves statutory effectiveness nor comprehensively measures substitution or harassment. It strengthens causal plausibility without closing the policy-evidence gap. (doi.org)

Advertising restrictions should be judged separately. DSA Article 28 prohibits profiling-based advertisements to reasonably known minors on covered platforms and does not oblige additional personal-data processing to establish minority. My inference is that restricting behavioral targeting removes one route for monetizing teen attention; contextual advertising means it does not eliminate engagement incentives. The additional normative justification is limiting commercial profiling itself, independently of whether recommendation quality improves. Neither justification should be mislabeled as demonstrated well-being improvement. (eur-lex.europa.eu)

B’s enforcement objection also changes implementation, not necessarily the adoption decision. NIST documents demographic and image-quality variation in facial age estimation; I would therefore require deployment-specific threshold-error audits and appeals, rather than mandatory facial estimation. The EDPB describes age-threshold tokens and requires proportionality, minimization and alternatives. These are design requirements, not proof that any deployed system is privacy-safe. (nist.gov)

I therefore favor phased legal adoption, separately calibrated for ages 13–15 and 16–17, preserving educational and support access. Require independent longitudinal follow-up across accounts, services and devices, including less-moderated platforms and AI companions; measure harmful exposure, repeat harassment, compulsive use, well-being, exclusion and privacy costs. Stagger limits, advertising restrictions, recommendation changes and contact protections separately. Pre-register revision or withdrawal criteria. This differs from B’s targeted-protection-first approach: adopt accountable limits alongside those protections, while requiring evidence of additional net benefit before retaining or expanding them.

B’s strongest position is that compulsory restrictions should demonstrate additional safety benefits over safer recommendations, contact controls and reporting before becoming the default; voluntary experiments and declining usage do not establish that benefit. Displacement, lost support access and age-assurance costs could outweigh gains, so independently evaluated targeted protections should precede broader restrictions.

My position remains stronger as a conditional policy judgment: adopt enforceable, phased restrictions alongside targeted protections because those interventions address different potential causes of harm, while making continuation contingent on incremental benefits. This is not evidence that the combined policy already works; it is a preference for reversible, mandatory implementation rather than voluntary testing alone.

Two findings justify that limited step without establishing the final outcome. Wang’s June 2026 difference-in-differences study estimates a 22% intensive-margin reduction in total Internet time following China’s gaming restrictions: displacement need not replace every prevented minute. Crucially, it also reports no clear physical-health gains and suggestive worsening of mental well-being amid pandemic confounding; reduced time cannot stand in for safety. (sciencedirect.com) Davis and Goldfield randomized 220 distressed participants aged 17–25 to requested social-media reduction or control; three weeks of restriction improved self-reported depression, anxiety, fear of missing out and sleep. That supports causal plausibility, not compulsory younger-teen effectiveness or lasting cross-platform benefits. (doi.org)

Advertising warrants an independent decision. DSA Article 28 prohibits profiling-based advertising to reasonably known minors on covered platforms without obliging additional age-determination data processing. (eur-lex.europa.eu) My economic inference is that removing this monetization channel could weaken attention-capture incentives, not eliminate them; my separate normative judgment favors limiting commercial profiling. Neither establishes reduced harassment.

B is right about the enforcement prerequisite: EDPB paragraph 21 treats effectiveness as a precondition for necessary and proportionate legally mandated age assurance. (edpb.europa.eu) I therefore require predeployment evidence of age-threshold accuracy, circumvention resistance, minimal retention, accessible alternatives and appeals—not merely a promised later audit; these requirements follow the statement’s effectiveness and safeguards framework. (edpb.europa.eu) Limits should distinguish ages 13–15 and 16–17 and preserve educational and support access.

I would change my mind after an independently governed, preregistered, adequately powered 12-month cluster-randomized or credible staggered-rollout study comparing targeted protections alone against those protections plus limits, advertising bans, or both. Follow the same teens across devices, alternative accounts and services—including AI companions—with consented measurement, attrition sensitivity analyses and independent exposure coding. Separately estimate recommendation and abuse-prevention changes; measure harmful exposure, repeat harassment, compulsive use, validated well-being, exclusion and privacy costs. If confidence intervals excluded prespecified meaningful benefits and documented burdens favored targeted protections alone, I would withdraw the additional restrictions as safety policy. A statistically inconclusive null would not suffice; reproducible net harm would.