Australia now has several early post-implementation numbers that can sound like answers to the same question. They are not.

There are millions of reported account-level actions. There is lower reported own-account holding among under-16s. There is a smaller movement in recent use of restricted platforms. There is no statistically supported change in a broader frequency measure. And eSafety has no baseline from which to calculate a change in time spent.

The useful question, then, is not which number is “the” result. It is which number is evidence of what.

One policy, several measures

Australia’s early evidence sits at different layers.

At one layer are platform account actions: removals, restrictions and other access controls. At another is account status: whether a child reports having their own profile or account. Then come behavioural measures: whether a restricted platform was used recently, how often social media was used more broadly, and how much time was spent using it.

These are not interchangeable. A platform can remove an account without proving that one child has stopped using social media. A child can stop holding their own account while continuing to access a platform. Recent use, frequency and time spent can also move differently.

The further a claim moves from account administration toward behaviour, the more carefully its metric has to be specified.

What millions of account actions actually count

eSafety reported that approximately 4.7 million age-restricted accounts had been removed or restricted by mid-December 2025. By early March 2026, platforms had reported more than 300,000 additional account-level actions preventing sign-in or access.

The unit is accounts, not children.

One person can hold multiple accounts across services. Some actioned accounts may have been inactive. Platforms also apply different kinds of controls.

The figures therefore establish large-scale administrative intervention. They do not establish how many unique Australian under-16s were affected, what share stopped using a service, or how many stopped using social media.

Own-account holding moved

eSafety’s published repeated-wave analysis follows 803 children aged 10–15. On its harmonised own-account measure, reported account holding moved from 52.4% at baseline to 42.1% at three months: a 10.3 percentage-point decline.

That is a statistically significant descriptive movement within the published cohort.

But the timing qualification matters. Some baseline fieldwork crossed the 10 December implementation boundary, and the public documentation does not fully resolve how that timing maps onto every participant in the analytic cohort. The comparison can therefore be used descriptively, not as a clean causal pre-policy/post-policy estimate.

The defensible claim is that reported own-account holding was lower at three months than at baseline within eSafety’s repeated cohort. It is not that the restrictions caused account holding to fall.

Account status moved more than recent use

In the same cohort, past-four-week use of at least one restricted platform moved from 85.9% to 81.5%: a 4.4 percentage-point decline.

Placed beside the account measure, the difference is clear:

  • own-account holding: 52.4% → 42.1%, down 10.3 percentage points;
  • recent restricted-platform use: 85.9% → 81.5%, down 4.4 percentage points.

Account status moved more than recent use.

eSafety’s data also show that some children reported using restricted platforms without holding their own account. That is enough to show why account holding is not an adequate proxy for actual recent use.

It does not explain the gap. The evidence does not establish how much of the difference reflects borrowed accounts, logged-out access, false age information, movement between services or any other form of adaptation.

The measurement distinction is firm. The mechanism is not.

“Use” still isn’t one behaviour

Recent platform use is only one behavioural measure.

eSafety also asked how often children used social media more broadly when they were not at school or work. Daily-or-more use moved from 60.6% at baseline to 57.7% at three months. The mean ordinal frequency score remained 3.9 at both waves. The change was not statistically supported.

That does not cancel the movement in recent restricted-platform use. The measures answer different questions: one captures whether named restricted platforms were used during a four-week window; the other captures broad social-media frequency.

Time spent is separate again. At three months, eSafety reported average self-reported use of 2 hours 9 minutes on school days and 3 hours 58 minutes on weekend days among the relevant respondents.

But there is no baseline time-spent measure in the public survey report. It therefore cannot show whether time spent rose, fell or stayed the same between baseline and three months.

A three-month level is not a longitudinal change. Nor should survey self-report be treated as equivalent to passive device observation.

An independent study does not resolve the uncertainty

An independent BMJ study provides a behavioural check, but not a direct replication of eSafety.

Its sample, age coverage, recruitment, lookback period, outcome definitions and analytical design differ. Its percentages should therefore not be merged with eSafety’s.

Among sampled 14–15-year-olds, daily social-media use moved descriptively from 78% before implementation to 69% at follow-up. The study also recorded substantial continued access to restricted platforms among under-16 participants.

Its formal regression-discontinuity analysis, however, did not find statistically supported age-16 discontinuities for either daily use or its ordinal time-use measure. The confidence intervals were wide, and the authors acknowledged power limitations.

That constrains strong claims that a clear behavioural effect has already been demonstrated at the legal threshold. It does not prove zero effect.

BMJ’s role is therefore to narrow interpretation in both directions: the data do not support a clean demonstrated behavioural effect, but neither do they support the claim that nothing changed.

Adaptation exists; its scale remains unclear

BMJ also documents continued access and adaptation among participants, including continued own-account access and other workarounds.

That establishes existence, not national prevalence.

The current evidence does not show how common circumvention is across all Australian under-16s, how much it explains the difference between account holding and recent use, or whether those behaviours will persist.

Adaptation is therefore part of the early picture, but not yet an established explanation for the account/use divergence.

What the early evidence can carry

The evidence is strongest at the administrative layer: platforms report large-scale account intervention.

At the next layer, eSafety’s repeated cohort shows lower reported own-account holding at three months than at baseline. The same cohort also shows lower recent restricted-platform use, but by a smaller amount.

Beyond that, the picture becomes more metric-dependent. Broad frequency did not show a statistically supported change. eSafety’s public survey cannot establish a longitudinal change in time spent. Independent behavioural evidence shows continued use, some descriptive movement and visible adaptation, but does not resolve causal magnitude.

Several things remain unestablished: how many unique children sit behind the account-action totals; what caused the observed account and use changes; how much adaptation explains the gap between them; how common circumvention is nationally; whether time spent changed in eSafety’s cohort; and whether these early patterns will persist.

The measures are not competing versions of the same answer. They describe different layers of the same system.

Any claim about the “impact” of Australia’s under-16 restrictions is therefore incomplete unless it names the metric.