Korean writing now has a sixth K-pop generation. We tested the first five and the lines are a year off

A Korean outlet described a group this week as a sixth-generation girl group. The phrase went by without explanation, the way it usually does. First generation, second, third — the labels are used as if some body had set them.

None has. No institution defines the boundaries, no database records them, and the years shift depending on who is writing. They are a convention, which is a perfectly reasonable thing for a convention to be. But a convention can be checked, so we checked it.

What we did

We took the 380 Korean acts from our member-versus-group count, asked Wikidata for each act’s founding year, and kept the 308 that had one. Set aside suspected sub-units — they read like people rather than groups, because the sub-unit’s own name is the part nobody knows — and 182 acts remain. Against each we put twelve months of English Wikipedia reading, act page and member pages added together.

Then we wrote the conventional boundaries into the code, in the open, so they could be argued with: first 1996–2002, second 2003–2011, third 2012–2017, fourth 2018–2022, fifth 2023–2025, sixth 2026 onward.

The medians climb. That part of the story holds

Later acts are read more. A group from the fourth generation is read about twice as much as one from the first, and the newest bin is higher still. If that is all you wanted from the labels, they deliver it.

But the bands sit on top of each other

A median is one point. What decides whether a label is useful is the spread, so we measured how much the middle half of each generation overlaps the middle half of the next.

At 86% to 100%, knowing an act’s generation tells you close to nothing about how much that act is read. Two of the pairs overlap completely: one generation’s middle half sits entirely inside the other’s. The label describes when a group started. It does not describe the group.

And the breaks are a year off the lines

This is the test we care about most, because it does not use our bins at all. Line the acts up by founding year, compare the median reading of each year against the next, and see where the big moves are. If the boundaries are real, the big moves should sit on them.

Sixteen year-to-year gaps had at least four acts on both sides — enough not to be one act moving. Two of those sixteen are generation boundaries. Among the two largest jumps, neither is a boundary.

The detail worth sitting with: in both boundary years we could measure, the real break is adjacent to the line rather than on it.

• 2012, the most-quoted boundary of all — the third generation, the BTS and Twice era — ranks 11th of 16 at 1.75×. The year before it, 2011, is the single largest break in the whole set at 9.06×.
• 2018, the fourth-generation line, does rank 5th at 3.36×. But 2019, immediately after it, is 8.58× — more than double.

So this is not a finding that nothing changes. Something changes, twice, hard. It changes a year from where the label puts it.

One column we got wrong, and what it turned into

The fifth generation shows a median member share of 2.5% — the share of an act’s reading that lands on member pages rather than the act page. Every older generation is above 46%. Our first explanation was that these members have no English Wikipedia articles yet.

That is false, and measuring it said so. Fifth-generation acts carry a median of 4 members with an article; the third generation carries 2. The articles exist. They are barely read.

Four of the five send under 5% to members. Whether that is the encyclopaedia lagging or attention on very new groups genuinely sitting on the group name, five acts and one twelve-month window cannot tell apart. So we claim neither. What we will not do is print 2.5% as a fact about how fans of new groups behave — and we print all five rows so nobody has to take the median on trust.

The sixth generation is in print before it is in the data

Our set holds no acts founded in 2026. That is not a claim that no such groups exist — Korean outlets are naming them now, and they are right that something is starting. It is a statement about what can be counted: an act founded this year has had no twelve-month window to be read in, and English Wikipedia has usually not written it down yet.

The label arrives first. The record arrives later. For a while there is nothing to count, and we would rather say that than produce a number from five months of a group’s existence.

What would change our mind

If the next recount moved the big breaks onto 2012 and 2018, we would say so. The year-by-year table is the part that does not depend on our bins, and it is on the data page for anyone who would draw the lines differently: do the generation labels show up in the reading.