RN
Two weeks before 2021 began, the platform that produces the numbers everyone quotes deleted most of its library.
On 14 December 2020 it suspended everything that had not come from a content partner or a member of its model programme. Contemporaneous reporting put the catalogue at roughly 13.5 million videos before and roughly 4.7 million after. The change followed New York Times reporting alleging the site hosted footage of minors and of non-consensual acts, and the withdrawal of Visa and Mastercard from processing for it.
That is the fact that governs every "most watched of 2021" list, including the one this article used to be, and it is why the list is not here.
Why one deletion invalidates a year of rankings
A popularity ranking is a comparison. It only means something if the thing being compared held still.
What people watched in 2021 was measured against a catalogue that had just lost about two-thirds of its items, weighted toward exactly the material least likely to have a verified uploader — old uploads, amateur re-posts, rips. The categories that survived best were the ones dominated by studios and verified performers, because those were the accounts the policy kept.
So a category that rose in 2021 may have risen because more people wanted it, or because its competitors had been removed. Nothing in a published rank distinguishes those, and the platform has no reason to.
The same applies to performer rankings that year, harder. A performer's apparent traffic depends on how many of their uploads still existed, which after December 2020 depended on whether the accounts that had posted them were verified. That is a fact about upload provenance, not about audiences.
Read the 2021 numbers as the first measurement of a substantially new library. They are a valid description of that library. They are not a continuation of the series that came before them, and treating them as a trend line is the error.
The gap the original was chasing
The article this replaces was written in late 2021 and opened by complaining that the annual demographic report had not appeared for 2020. That complaint was accurate. The series has run since 2013 and the 2020 edition is the one that is missing — which, given what the company spent that December doing, is not mysterious.
It resumed afterwards and has continued to appear each year since. So the archive is not closed. It just has a hole in it at precisely the point where the platform changed shape, which means the discontinuity is invisible if you only read the reports.
What a platform report can and cannot tell you
These documents get quoted as though they described human sexuality. They describe one company's search box.
They measure typing, not watching. A search term counts when someone enters it. What they then watched, for how long, or whether they found it, is not in the figure.
They measure that platform's population. People who use paysites, direct-to-fan platforms, forums or file hosts are not in the data, and those populations are not random samples of each other.
The vocabulary is the platform's. Search terms are constrained by the tag taxonomy the site offers and by what its autocomplete suggests. A category that gets a prominent tag accumulates searches partly because it is easy to search for.
Search volume includes curiosity. A term spikes when a name is in the news. Some of that traffic is people checking whether something exists.
Geography is now unreliable. Following state-level age-verification statutes, the platform is unavailable in a large share of US states, and in France. Users in blocked regions who reach it through a VPN are counted wherever the exit node sits. Any recent country- or state-level comparison is measuring that as much as anything else.
None of this makes the reports worthless. It makes them evidence about a specific system, which is a smaller and more interesting claim than the one the headlines make.
Reading them usefully
The parts that survive scrutiny are the relative ones, within a single edition, where the same instrument measured both sides.
Comparisons between demographic groups in the same year are meaningful — the same search box, the same taxonomy, the same window. So are shifts in the language people use for a category, which tell you something real about how a thing is being talked about even when the volumes are noisy. So are step changes large enough that no methodological quibble accounts for them.
What does not survive is the ranked list of titles, performers and categories, quoted across years as though a stable population were being measured with a stable instrument against a stable catalogue. In this particular series, at this particular point, none of those three held.
That is a duller finding than a top ten. It is also the only one the evidence supports.