Korean pop culture, in numbers

Stars

A piece asked if K-pop's new power fans are middle-aged. We can't measure age, so we tested the proxy

Wikimedia publishes no reader age. It does publish device, which is often used as a proxy for one. Across 372 Korean acts, desktop share barely moves with debut era: 2.9 points from the 1990s to the 2020s.

The table this is written from: /desktop-or-phone

#kpop#wikipedia#audience#device#method#attention

An English-language piece this week put a question that is genuinely interesting: are K-pop’s new power fans middle-aged Americans with money and time to spend?

We cannot answer it. Wikimedia publishes no reader age, and nothing below is a claim about how old anybody is. But there is a reason to write this anyway, because device is the thing people reach for when they want to talk about age without measuring it — phones for the young, desktops for everyone else. Wikimedia does publish device. So we tested the proxy.

What the split actually is

For the twelve months to August 2026 we counted English Wikipedia reading for 372 Korean acts twice over: once for desktop, once for everything. Phone is the remainder — mobile web and the app together, which the data cannot separate.

What we counted Value
Acts measured 372 of 380
Reads in the window 28,363,111
Desktop, all reading added together 32.4%
Desktop for the median act 34.2%
Phone (mobile web plus app) 67.6%

Two numbers rather than one, because they answer different questions. Added together, the biggest acts dominate; the median describes a typical act. The 1.8-point gap between them says the very largest acts sit slightly further toward the phone than the rest.

It does not move with debut era

If device tracked audience age at all, an act that debuted in 1997 should be read differently from one that debuted in 2023. Here is every ten-year band holding at least ten acts.

Founded Acts Desktop share, median act
1990s 30 34.2%
2000s 62 37.0%
2010s 173 34.1%
2020s 34 35.7%

The entire spread is 2.9 points, and it does not even run in one direction — the 2000s sit highest and the 2010s lowest, with the newest band in between. Across three decades of debuts, that is flat.

We tried to break it, and could not

A percentage of a small number is fragile, so the obvious objection is that the flatness comes from hundreds of little acts whose shares wobble. We removed them: only acts read at least 100,000 times in the window, 61 of them.

Founded Acts Desktop share, median act
2000s 13 34.2%
2010s 24 32.5%
2020s 14 35.3%

Still flat — 2.8 points, median 33.5% across all 61. This is where the finding should have fallen apart, and it held.

So what can be said

Only this, and it is worth having: device is not a stand-in for the age of a K-pop audience. Anyone reading a desktop share as evidence about who a fandom is made of is reading a number that does not vary with the thing they want it to vary with.

Whether the original claim is true, we do not know. It may well be. What we can say is that the most easily available proxy for it does not support it, does not contradict it, and should not be cited either way.

Individual acts do sit far apart — the ends of our list run from 3.2% desktop to 57.9% — but read the reads column before building anything on that. The extremes are mostly acts with small totals, which is exactly where a share stops meaning much. We print them anyway, on the data page, with the totals beside them: desktop or phone, and why it is not age.

Where these numbers come from

Sources

  • Wikimedia Foundation — Pageviews REST API — en.wikipedia, agent=user, monthly, 2025-09 to 2026-08. Fetched twice per act: access=desktop and access=all-access. Phone is the remainder, so it is mobile web plus the app together. · https://wikimedia.org/api/rest_v1/
  • Wikidata — SPARQL — property P571 (inception), used only to place each act in a ten-year band. · https://query.wikidata.org/sparql

What we checked

  • The denominator was verified against a count made a day earlier for a different question. Desktop share is desktop reads over all-access reads, so a wrong denominator would produce a wrong share silently. We already held all-access totals for these acts from the member-versus-group collection, fetched them again today, and compared: 372 of 372 acts agree to within one percent, with a median difference of 0.0%. The well did not move under us.
  • The flat result was tested a second time with the noisy acts removed. A percentage of a small number swings easily, so we repeated the comparison using only the 61 acts read at least 100,000 times in the window. The spread between decade bands stayed inside 2.8 points and the median moved by less than one point. If the flatness had come from small acts, this is where it would have broken.
  • Eight acts were dropped rather than shown. Our floor is 1,000 reads in the window; below that a share moves on a handful of visits. The dropped acts are counted and named in the data file, not filled in as zero.

The data behind this

Written from the same data

If you work in this business

Everything on this page is measured from the same weekly Netflix lists, per market and per title. We publish the pages free; what we sell is the same measurement cut to one company's catalogue — including the figures behind /desktop-or-phone.

What a company sheet contains, and what it cannot tell you →

← All 215 articlesFront page