AdultDVDTalk has run geo-based age verification for a while now — visitors get routed into a check only in jurisdictions that require one, everyone else sees the site exactly as they always have. That gives a clean dataset for the regions where verification is actually happening, without diluting it with traffic that never touched a gate at all.
The assumption most people bring to this topic is that the friction lives in the check itself: people failing, getting rejected, giving up mid-process. A year of real session data says otherwise.
Where people actually go
Year to date, 188,029 sessions hit an age gate on ADT. 143,858 of them — 76.5% — never resulted in a verified user. That number alone sounds like an indictment of the process. It isn't, once you look at where the abandonment actually happens:
- 64,898 people (34.5%) never even load the verification page.
- 74,023 (39.4%) load it and never start.
- 496 abandon partway through the liveness check.
- 4,441 abandon during document review.
Add it up and 96.6% of all abandonment happens before anyone attempts verification. The actual check — the part everyone assumes is the obstacle — accounts for well under 4% of the people who leave. Almost nobody who's going to bail does so because the process was hard. They decide before they try.
What happens once someone actually tries
For the people who do start, the process works. Liveness checks succeed 96.3% of the time across 50,000+ attempts. Failures skew overwhelmingly technical, not adversarial: low-confidence liveness reads (bad lighting, a bad angle) and unreadable documents account for the vast majority of failures. Fake IDs, underage attempts, and multiple-face detections combined make up about 1.3% of all failures — everything else is someone who had a legitimate shot and hit a quality problem, not a fraud problem. On the document side specifically, most failures are DOB extraction issues — an OCR problem, not a "this person shouldn't pass" problem.
None of that shows up in the 76.5% headline number, and it should. The check itself isn't the friction. The decision to open it is.
The design of the requirement matters more than whether one exists
Not every jurisdiction that requires age verification requires the same thing. Some treat a passed liveness check as sufficient on its own, falling back to document verification only if the biometric check is inconclusive. Among jurisdictions where document verification is the norm, we apply a calibrated liveness-confidence threshold in a subset of them as an alternative path — a sufficiently strong liveness match, combined with a low estimated minor probability, can clear verification without a document step.
Grouping this year's regional data by which of those three designs applied produces a pattern worth taking seriously:
| Compliance design | Regions | Sessions | Verified | Completion rate |
|---|---|---|---|---|
| ID required, no skip option | 14 US states | 19,824 | 1,478 | 7.5% |
| ID required, skippable if liveness is confident | 7 US states | 39,503 | 7,992 | 20.2% |
| ID not required (liveness-first, document is a fallback) | AU, DE, FR, GB, IT | 114,836 | 28,397 | 24.7% |
That's roughly a 3x spread in completion rate between the most restrictive design and the most permissive one — not between "has AV" and "doesn't," but between different implementations of AV that are all legally satisfying an age-verification mandate. The shape of the requirement is doing as much work as the requirement itself.
It's not the whole story. Within the least restrictive tier, completion still ranges widely — Germany converts at 13.2%, the UK at 38.4% — so something beyond design (likely regional trust in handing over biometric data at all) is also at play. But a 3x swing from implementation choice alone is a large enough lever that it deserves more attention than it gets in most conversations about AV, which tend to treat it as a single yes/no question rather than a design space.
When the economics don't work at all
Not every jurisdiction stays served. Tennessee's requirement includes reverification on a roughly hourly cadence — meaningfully more frequent than a one-time or daily check. Meeting that would mean either re-running liveness checks on that cadence or engineering a shorter-lived verification state just for that state, and at the traffic volumes involved, the ongoing cost doesn't clear the bar. So Tennessee is blocked outright rather than served badly or worked around. That's the same underlying calculus as everything above, just at the extreme end: past a certain point, a compliance design doesn't just add friction, it stops making sense to build for at all.
What this actually says
The public debate around age verification is mostly framed as should this exist. The data here suggests the more consequential question is what shape it takes once it does. Almost everyone who's going to leave an age-gated site leaves before attempting anything, the check itself performs well for the people who try it, and the single biggest lever over completion isn't the presence of a law — it's whether the implementation gives someone a reasonable path through without demanding more than the moment calls for.