Fifteen historical cases across media, gambling, alcohol, credit, gaming and driving. The restrictions that worked share three conditions, and most current age-gating proposals satisfy two of them.
Companies and governments increasingly answer questions about youth access the same way: an age gate, a parental approval step, an identity check, or a separate product for minors.
The assumption underneath is that stronger restrictions produce different behavior. The historical record is less obliging. Some restrictions changed almost nothing. Some changed who completed the transaction without changing whether it happened. A small number genuinely suppressed the activity, and those cases have a specific structure worth understanding before designing anything.
This study looks at what happened after the restriction, measured wherever possible with transaction, enforcement or panel data rather than with what young people said about themselves.
Cases were selected on evidence quality, not on category coverage. Each one required demand from young people before the restriction, an age-based intervention with real enforcement behind it rather than a warning label, and enough post-implementation measurement to distinguish compliance from circumvention.
The cohort spans social platforms, rated media, gambling, alcohol, tobacco, consumer credit, youth payments, app purchases, school networks and novice driving. It deliberately includes physical-world restrictions where substitutes are scarce, because a cohort built only from digital cases would answer a narrower question than the one being asked.
| Case | Year | Enforcement layer | Adult role | Primary outcome | Confidence |
|---|---|---|---|---|---|
| Minimum drinking age 21 (US) | 1984– | Point of sale | Supplier | Compliance at the gate, leakage through social supply | Fact |
| Graduated driver licensing | 1996– | Government / identity | Enforcer | Compliance | Fact |
| COPPA | 1998/2000 | Product | Circumvention agent | Parental delegation, then circumvention | Reported |
| V-chip | 2000 | Device, parent-activated | Non-adopter | No material change | Fact |
| FTC rated-media mystery shopping | 2000–2013 | Retail | Purchaser | Compliance at the gate | Fact |
| CIPA school filtering | 2000– | Device and network | None at school | Insufficient evidence | Unresolved |
| CARD Act Title III | 2010 | Payment / identity | Reluctant cosigner | Compliance; delayed credit accumulation | Fact |
| South Korea shutdown law | 2011–2021 | Identity | Credential supplier | Circumvention, then repeal | Reported |
| Tobacco 21 | 2016–2019 | Point of sale | Secondary supplier | Mixed; self-report and biomarkers diverge | Unresolved |
| China minors' gaming limits | 2019, 2021 | Government infrastructure | Credential supplier | Circumvention with partial compliance | Reported |
| US adult-site age verification | 2023– | Product with identity check | None | Substitution | Fact |
| UK Online Safety Act age checks | 2025 | Product with identity check | None | Substitution | Fact |
| Australia under-16 social ban | 2025– | Product | Awareness declined | Little material change so far | Fact |
| Youth versions of adult products | 2015– | Product | Permission granter | Product adaptation, weak demand | Reported |
| UK regulated gambling age limits | 2005– | Licensed operator | Present at the activity | Gate holds, activity relocates | Fact |
Confidence reflects the strength of the underlying evidence, not the size of the effect. Fact means primary regulator, court or panel data. Reported means credible secondary evidence. Unresolved means the available measurements cannot separate the competing explanations.
Most discussion of age verification collapses two different questions into one. They need to stay separate.
The first is whether the restriction stopped the young person from completing the action themselves. The second is whether it stopped them from ending up with the product, the experience or the money. A restriction can score very high on the first and near zero on the second.
Rated media is the cleanest illustration because the enforcement is voluntary and there is no statute behind the rating. In the Federal Trade Commission's 2012 undercover shop, only 13 percent of unaccompanied 13-to-16-year-olds were able to buy an M-rated video game, and 24 percent could buy a ticket to an R-rated film, a program low. Retail compliance of that order is unusual in any regulated category.
It does not follow that teenagers stopped playing M-rated games. The gate governed who could stand at the counter. An adult could buy the identical product minutes later.
The upper-right cell holds most of the cohort. High gate effectiveness with low access suppression is the normal outcome of an age restriction, not the exception. The lower-left cell is empty because no case was found where a weak gate nonetheless suppressed access.
Across the cohort, the adult who is meant to hold the line is more often the mechanism by which the line is crossed. This is not a claim that parents are careless. Their objective is simply different from the regulator's. A platform sees an age restriction. A parent sees a child asking for something the parent considers reasonable.
The most direct measurement comes from gambling, where the age limits are statutory and licensed operators face test purchasing. In the Gambling Commission's 2025 survey of 3,666 eleven-to-seventeen-year-olds, 75 percent of young people who gamble do so with a parent, carer or guardian, and 8 percent gamble alone. Thirty percent had spent their own money in the previous year, up from 27 percent, and the regulator attributes the increase to legal or unregulated forms such as private betting between friends rather than to underage access to licensed products.
Alcohol shows the same shape from a different angle. The minimum drinking age produces one of the sharpest behavioral discontinuities in economics: drinking days rise 21 percent at the twenty-first birthday, with a 9 percent jump in mortality, which can only happen if consumption was genuinely suppressed beforehand. Yet most underage drinking is supplied by someone older rather than purchased directly.
The precise finding is narrower than the headline. Adult intermediation is common when the action is cheap and transferable. It is not universal, and the next section is about where it stops.
The cohort separates cleanly on one variable, and it is not enforcement strength. It is what saying yes costs the adult.
The permission price is what an adult takes on when approving or completing a young person's action, counting effort, financial exposure and future commitment.
Where the permission price is near zero, intermediation clears almost everything. A child selects a $4.99 item, a parent recognizes the merchant and taps approve. Where the price includes durable liability, it stops clearing.
One transaction, recognized merchant, no continuing obligation. Almost all demand passes through.
A trip to the counter. Enough friction to shift timing, rarely enough to stop the purchase.
Sharing an account or an identity document. Common where enforcement is strict and the good is wanted badly.
Cosigning a revolving line. Exposure to debt the adult does not control, for years. Most decline.
Position on this scale predicts outcomes better than the strength of the enforcement behind the restriction.
Title III of the CARD Act supplies the cleanest test. From 2010, consumers under 21 needed either an independent ability to repay or an adult cosigner to obtain a credit card. It also restricted acquisition, limiting campus marketing and prescreened offers to under-21s.
Using the New York Fed Consumer Credit Panel, researchers found that twenty-year-olds became 8 percentage points less likely to hold a credit card, a 15 percent decline, while cosigned cards rose by roughly 3 percentage points among those who still held one. The authors treat the first figure as a lower bound, since the Act itself reduced how many young people appeared in credit bureau data at all.
Adults were available. The same research shows that the young people affected disproportionately had creditworthy, engaged parents. Most of those parents did not cosign. Adult intermediation absorbed a minority of the displaced demand, and that is the only case in the cohort where it did.
The practical version: an approval flow can be scored. As the permission price rises, a smaller share of intended transactions survives the adult layer.
The CARD Act restricted two different things at once, and separating them explains most of the cohort.
The first is purchasing power. The median combined credit limit for eighteen-to-twenty-year-olds was about $1,000. Removing that line did not remove much consumption, because cash, debit, family money and other rails were all still there.
The second is time inside the credit system, and nobody could supply that on a young person's behalf. The same research estimates that each additional quarter of credit-card experience is associated with roughly 2.4 additional credit-score points, so a delay of several years leaves a measurable gap that persists into the period when someone applies for larger financial products.
The value is delivered at the moment of the transaction, and any authorized party can complete it.
The value comes from the young person having personally participated over a period of time, and a proxy cannot generate it for them.
Only the credit history case is established by the cohort. The other three share the same structural property, which is a reasonable basis for expectation and not a measured result. They are labelled as inference for that reason.
Graduated driver licensing sits on the same side of the divide and explains why it is the strongest suppression case in the cohort. Pooled estimates put the crash rate ratio for sixteen-year-olds at 0.84, and night-driving restrictions cut injury and fatality rates in restricted hours by roughly a quarter. No adult can drive on a minor's behalf. The activity itself is non-transferable, so there is no intermediation channel to absorb the demand.
The natural assumption is that outcomes improve as enforcement moves deeper into the stack, from the product to the operating system to identity and payments. The cohort does not support that as a rule.
China runs the strictest identity infrastructure in the set, limiting minors to three hours of online gaming a week against a state ID database with facial recognition on suspect logins. It produced the most inventive circumvention in the cohort. People's Daily reported that Tencent sued more than twenty e-commerce sites within a week of the 2021 rules as listings appeared renting verified adult accounts to minors, and minors reportedly borrowed relatives' credentials and photographed adults to defeat facial checks. Both claims rest on secondary reporting and are coded accordingly.
At the other end, restrictions that leave one provider compliant and another not simply move the traffic. A pre-registered synthetic control study of US state age-verification laws found searches for the leading compliant platform down 51 percent three months after passage, with the leading non-compliant platform up 48.1 percent and VPN searches up 23.6 percent. Separate work estimates a net reduction of roughly ten hours per hundred previously spent. Real, and much smaller than the traffic decline on the compliant site implies.
Depth helps only when the layer is genuinely unavoidable. School network filtering sits at the device layer and is among the leakiest cases here, because the school device is one of several a student carries. An app store is a real chokepoint for native apps and none at all for the browser.
There is an observable trend since 2023 of regulators moving from product-level restrictions toward identity and device-level requirements after product-level rules underperform. One cycle is not a law. Alcohol enforcement has sat at the point of sale for forty years without migrating anywhere.
Age restrictions change behavior most reliably when three conditions hold together: the young person must personally pass the gate, the gatekeeper carries meaningful incentive or liability to enforce it, and the action cannot be cheaply completed by an adult or routed to a substitute rail.
When those conditions fail, restrictions tend to change who executes the action, or where it happens, more than whether the underlying demand survives.
Consumption is the easiest thing to transfer through an adult. Accumulation is the hardest, because the value has to belong to the individual and build over time.
Two conditions are satisfied by most current age-gating proposals. The third rarely is.
Everything above is history. The following are current, which makes them tests of the pattern rather than evidence for it.
By December 2027, under-16 access to covered social platforms in Australia will return to within two percentage points of the 85.9 percent pre-restriction baseline.
The restrictions took effect in December 2025. Three months later the regulator found that 81.5 percent of under-16s were still using an age-restricted platform, against 85.9 percent before the law, while account ownership fell from 52.4 percent to 42.1 percent. The gap between those two movements is the gate-versus-outcome distinction appearing in real time.
The report attributes continued access to a low barrier rather than to sophisticated workarounds, notes early signals of movement toward platforms outside the restricted set, and records a 10 percent increase in the share of parents who did not know their child had used social media. eSafety cautions that substantial change was not necessarily expected this early.
If access instead keeps falling as enforcement improves, the historical pattern is weaker than this study concludes.
By the end of 2028, a major payment network or processor will market a minor-status or age signal to digital platforms on a per-transaction basis rather than treating age solely as a one-time identity-verification event.
Identity can be borrowed once. Money has to move every time.
A credential is checked at signup and then trusted indefinitely, which is why borrowed credentials are the most common circumvention route in the cohort. A payment has to clear on every transaction, under a real name, leaving a record each time.
There are already documented cases of sportsbooks catching underage users at the deposit rather than at identity verification. In one, a minor betting on a parent's credentials was flagged when the funding account name did not match the registered account. Whether that route is the common one is not established. What it shows is that the payment rail carries an age signal the identity check had already missed.
The settlement gives 13-to-17-year-olds a materially different product: default time limits, overnight restrictions, school-hour notification controls, stronger age assurance and expanded parental controls. It is subject to court approval, and it is included here as an indication of where the industry is heading rather than as evidence for any of the findings above.
The part most relevant to this study is that it constrains discovery as well as execution. Meta has already limited advertiser targeting of teenagers to a narrow set of signals, and the settlement compresses when and how content reaches a minor at all. Companies reading this as a checkout problem will be solving the second constraint while the first one moves underneath them.
The findings reduce to two operating questions. They are different questions with different answers, and most product reviews only ask the first.
Find every point where a young person's intent requires someone else's authority, then price that approval in effort, money, liability and future commitment. The cohort says demand clears when the price is near zero and stops clearing well before the price becomes prohibitive.
A flow that hands a parent a bare product link and asks them to reconstruct the purchase is expensive. A flow where the young person configures everything and the adult receives a single request showing merchant, item, price and payment source is cheap. Same purchase, different completion rate.
Ask which parts of the product create value that has to accrue to the individual over time, then check whether the supervised version begins accruing it or defers all of it to the day they age into the full product.
Two young people can have an identical spending experience and arrive at adulthood in different positions, depending on whose name the underlying relationship sits in. The consumption is the same. The accumulation is not.
Restrictions on consumption tend to redirect the transaction, and the cost of getting them wrong is abandonment. Restrictions on accumulation compound quietly, and the cost of getting them wrong shows up years later in someone else's product.