TikTok paper captures >99% of an hour, pegs 269M posts/day

A near-complete slice of TikTok is rare, because the company does not publish granular posting volumes. A new arxiv paper from Benjamin Steel, Miriam Schirmer, Derek Ruths and Juergen Pfeffer sets out an ID-sampling method they say captures more than 99% of posts from any given time window, then applies it to a full hour plus one-minute samples from every hour of a single day.

The headline numbers matter because they are researcher-derived, not platform-audited. The authors estimate 269 million posts were produced on the day they sampled. They also report that 18% of videos on the platform feature children, and that at least 0.5% of posts contain AI-generated content. If those figures hold up, they give regulators, advertisers, and rival researchers a public denominator for a platform that rarely offers one.

The abstract is thin on classifier detail. It does not say how 'features children' or 'AI-generated' were labeled, whether human coders or models drew the lines, or which specific day was captured. A one-day snapshot also cannot speak to seasonality or the aftermath of a moderation policy shift. The 0.5% AI figure is explicitly framed as a floor ('at least'), which suggests the authors themselves treat it as an undercount.

The method itself may prove more durable than any single statistic. If it replicates, independent teams get a way to spot-check TikTok's own transparency reporting, and the child-featuring share is the kind of number that tends to enter policy debates quickly. TikTok can close the door by changing its post-ID scheme, but until it does, a near-complete hour of the platform is now something anyone with the paper in hand can go collect.