What happened to Koishi Komeiji's vote structure in the 21st and 22nd Touhou Project Popularity Polls?

A data-driven look at the 2025 first-choice spike, the 2026 follow-up, and what the public data can - and cannot - tell us.

A note before we start​

This is not an accusation of cheating.

I am using words such as outlier, structural break, spike, and reversion in a statistical sense. A voting pattern can be highly unusual without being fraudulent, and aggregate data cannot identify the person or mechanism that produced it.

The purpose of this article is narrower:

Something unusual happened to the structure of Koishi Komeiji's support in the 21st Touhou Project Popularity Poll. What exactly changed, how unusual was it, and what did the 22nd poll tell us one year later?

Click to expand...

The public results are surprisingly good at answering the first two questions. They are much weaker at answering the third question - why.

1. The key distinction: "how many people voted for her" is not the same as "how strongly they ranked her"​

The character division uses weighted preferences. In the 21st and 22nd polls, the scoring system is:

• 1st choice: 3 points
• 2nd choice: 2 points
• other selected characters: 1 point

That means a character's final score is not the same thing as the number of people who selected that character.

For this analysis I use four basic quantities:

• N = actual number of voters who selected the character
• F = number of voters who placed the character 1st
• Support rate = N / total character-division voters
• First-choice share = F / N

The last measure is especially useful. It answers a simple question:

Among everyone who selected this character, what percentage treated them as their No.1 choice?

Click to expand...

This article is therefore not just asking whether Koishi gained popularity. It is asking what kind of support she gained.

2. The three-year picture​

All figures below are clickable; click an image to open the full-resolution version.

Here is the simplest overview.





At first glance, this is a story of a very popular character becoming even more popular. Koishi was already a long-term top character before 2025, and she had already won a previous edition of the poll.

But once first-choice and non-first-choice votes are separated, the 20th -> 21st transition looks very different from the 21st -> 22nd transition.

3. The strange part of the 21st poll was not simply that Koishi gained votes​

Between the 20th and 21st polls, total character-division participation rose from 56,855 to 58,509, an increase of only 2.91%.

Koishi's actual number of supporters rose from 12,939 to 14,815:

+1,876 supporters (+14.50%).

That is strong growth, but not yet the most interesting part.

Now split the growth by preference level:

• First-choice: 3,334 -> 5,133 = +1,799 (+53.96%)
• Non-first-choice: 9,605 -> 9,682 = +77 (+0.80%)

In other words, across the two aggregate snapshots, almost the entire net increase in Koishi support appeared in the first-choice count, while non-first-choice support was almost flat.

A wording warning is important here. This does not mean that "96% of new Koishi voters put her first." We do not have voter-level panel data, so we cannot track the same individuals from one year to the next. Some voters entered, some left, and some may simply have changed their ranking.

The defensible claim is:

At the aggregate level, roughly 96% of the net increase in Koishi support between the 20th and 21st polls was reflected in the first-choice count.

Click to expand...

That is the first reason the 21st-poll data deserve attention.

4. The same pattern appears in both raw counts and normalized rates​

A natural objection is that Koishi is already one of the largest characters in the poll. Large characters can produce large absolute changes.

So I looked at the Top 100 in two different spaces.

4.1 Raw-count space​

The horizontal axis below is the change in non-first-choice supporters. The vertical axis is the change in first-choice votes.

Koishi is approximately:

(+77, +1,799)



The visual point is straightforward: Koishi barely moves horizontally but shoots upward far beyond the main cloud.

This is not what a simple, broad popularity increase usually looks like. A broad increase would normally push a character to the right as well as upward.

4.2 Normalized rate space​

Now replace raw counts with:

• change in support rate
• change in first-choice share

Koishi becomes approximately:

• Support rate: +2.56 percentage points
• First-choice share: +8.88 percentage points

The outlier does not disappear after expressing both dimensions as normalized rates.

That is important. The 2025 pattern was not merely "Koishi is big, therefore her raw numbers move a lot." The composition of her support changed unusually quickly.

5. The historical comparison is just as striking​

I also compared annual changes in first-choice share across a fixed panel of long-running top characters.

Across 96 prior year-to-year changes in that reference panel, the largest earlier positive increase was about +3.98 percentage points.

Koishi in the 21st poll:

+8.88 percentage points.

That is more than twice the previous maximum in the predefined comparison set.

This does not mean there is a "98.97% probability" of wrongdoing. The empirical tail count is not a conventional p-value, and it certainly is not a probability of cheating.

The correct statement is much simpler:

Within this historical comparison set, no previous annual increase was as large as Koishi's 2025 jump.

Click to expand...

6. Koishi had already been trending upward - which makes the question more specific, not less important​

It would be misleading to pretend Koishi had been flat before 2025.

Her first-choice share from 2016 through 2024 was approximately:

22.94 -> 21.01 -> 22.26 -> 20.78 -> 21.87 -> 22.61 -> 23.16 -> 24.68 -> 25.77

So yes: there was already a late-period upward trend.

The question is not whether Koishi was rising. She clearly was.

The question is whether that trend explains a one-year jump to 34.65%.

Simple trend benchmarks also fail to account for the size of the jump. An OLS trend fitted to 2016-2024 predicts a 2025 first-choice share of about 24.95%, with a small-sample 95% prediction interval of approximately 21.51%-28.39%. Restricting the trend to 2020-2024 gives a prediction of about 26.58%, with a 95% prediction interval of approximately 25.13%-28.03%. A Theil-Sen robust trend estimate is also only about 25.45%.

These models should be treated as descriptive rather than as a literal data-generating process, but all of them place the observed 34.65% well above the continuation of the previous trend.



The fairest summary is therefore:

Koishi was already rising, but the 21st poll looks like an acceleration or structural break rather than a routine continuation of the earlier trend.

Click to expand...

7. Why the 3-2-1 system matters - and why it does not explain the spike by itself​

The 21st poll is also interesting because the final ranking was unusually sensitive to preference order.

In the same 21st-poll ballots:

• By raw number of supporters, Koishi was 3rd.
• Under a reconstruction similar to the older selection-plus-favorite scoring logic, Koishi was also 3rd.
• By first-choice votes, Koishi was 1st.
• Under the official 3-2-1 score, Koishi was 1st.

Her official score was 27,688, only 135 points ahead of Marisa.

This does not mean the new rules "created" the 34.65% first-choice share. A scoring rule cannot mechanically force voters to rank one particular character first.

What the rules did was different:

They amplified the electoral effect of an already unusually concentrated first-choice structure.

Click to expand...

That distinction matters. The scoring system explains why the concentration mattered so much for the championship. It does not explain why the concentration appeared.

8. A plausible normal explanation: Koishi's supporters are younger and newer​

There is real evidence for a demographic explanation.

In the 21st-poll questionnaire, Koishi supporters were notably younger than the full survey sample.



Koishi supporters were also more likely to say that this was their first Touhou popularity poll:

• Koishi supporters: 43.13%
• Overall questionnaire: 37.96%
• Difference: +5.17 pp

This is important evidence against simplistic narratives that treat Koishi's support as somehow detached from a real fandom base.

However, the questionnaire has a major identification problem.

We know the demographics of people who selected Koishi.

We do not know the demographics of the people who placed Koishi first.

Those are not the same conditional distribution.

So a younger/newer voter cohort is a plausible and important explanation, but the public questionnaire cannot tell us whether it specifically generated the +1,799 first-choice increase.

9. Comment data do not show a collapse in visible engagement​

Another common intuition is that if a large number of votes were low-engagement or artificial, perhaps they would not be accompanied by comments.

Koishi's comment counts were:

• 20th poll: 3,462
• 21st poll: 4,238
• 22nd poll: 4,612

The rough comment-to-support ratio did not collapse during the first-choice spike.

That weakens a very crude model in which a large mass of otherwise disengaged ballots simply appears without normal participation behavior.

But comments are only an auxiliary proxy. They cannot distinguish genuine fans from coordinated campaigning, strategic voters, manually operated multiple accounts, or other mechanisms.

So I treat comments as negative evidence against one simple model, not as proof of legitimacy.

10. Interpreting the ecosystem evidence more carefully​

A simple reading of the 2025 results might suggest that Koishi's surge was disconnected from related characters, songs, and works. The detailed conditional results show that this interpretation is too strong.

Koishi voters are very strongly connected to the expected ecosystem.

Among Koishi supporters:​

So the correct claim is not that Koishi voters stopped behaving like Koishi / Subterranean Animism fans.

The better description is:

The year-over-year growth rates were not synchronized.

Click to expand...

Koishi's first-choice count jumped by 53.96% in 2025, while Satori, Hartmann, Last Remote, and Subterranean Animism did not experience a comparable same-year first-choice surge.

That is a much narrower claim, but also a much safer one.

11. Co-voting data suggest mainstreaming rather than a tighter closed bloc​

Using the official co-voting tables compiled and discussed in reference [7], I recalculated standardized association measures including Lift and the phi coefficient.

Two of Koishi's strongest traditional relationships are especially interesting.

Koishi x Satori​

• Lift: 2.30 -> 2.18
• phi: 0.299 -> 0.291

Koishi x Flandre​

• Lift: 1.85 -> 1.68
• phi: 0.260 -> 0.219

The raw number of voters selecting both characters increased, yet the standardized relationship weakened.

This is not contradictory. Satori and Flandre themselves became more common choices across the whole electorate.

So Koishi voters remained highly likely to select them, but that behavior became less distinctive relative to the average voter.

The 22nd-poll conditional tables continue the same broad pattern: absolute co-voting remains strong, while the "specialness" of those links tends to decline.

That looks more like a support network becoming mainstream than a traditional small circle becoming increasingly closed and concentrated.

A technical note: the published co-voting table contains some pairwise inconsistencies where A-to-B and B-to-A common-voter counts are not identical. Because of these internal inconsistencies, exact pairwise metrics should be interpreted cautiously. Interval-based or sensitivity analyses are preferable to automatically choosing the larger of the two reported counts.

12. The 22nd poll gave us something much more valuable than another model: a new year of data​

Before the 22nd results were released, I had already used the following heuristic interpretation bands in my working notes. These were not a formal public preregistration, so they should not be treated as one; I mention them only to make clear that the interpretation of the 22nd result was not chosen solely after seeing the outcome:

• 25-29%: strong reversion
• 29-32%: partial reversion
• 32-36%: high persistence
• above 36%: further strengthening

I also noted that if Koishi's total support continued to grow while first-choice share fell, that would be compatible with a mixture of long-term genuine popularity growth and a special 2025 first-choice concentration.

The 22nd poll produced:

31.67%.

That falls directly into the previously defined working-note partial reversion range.

This is why the 22nd poll is so useful. It is not another statistical model fitted to the same 2025 observation. It is a genuinely later observation that did not exist when the original anomaly was identified.

13. The 2026 growth pattern was almost the mirror image of 2025​

This is probably the most intuitive result in the entire follow-up.

20th -> 21st​

• Total Koishi support: +1,876
• First-choice: +1,799
• Non-first-choice: +77

21st -> 22nd​

• Total Koishi support: +1,482
• First-choice: +29
• Non-first-choice: +1,453

The two transitions are almost mirror images.

In 2025, support became sharply more top-heavy.

In 2026, first-choice support almost stopped growing, while broader support caught up.

I find "spiking -> broadening" a much more useful description than arguing immediately about whether the cause was legitimate or illegitimate.

14. The Top-100 distributions confirm that the 2025 pattern did not repeat​

In raw-count space, Koishi's 21st -> 22nd position is approximately:

(+1,453 non-first-choice, +29 first-choice)



This is almost the geometric opposite of the previous year's vertical spike.

However, raw counts need one more correction. Total character turnout increased by 10.16% in the 22nd poll.

Koishi's supporter count increased by 10.00% - almost exactly in line with the whole division.

After normalization:

• Support rate: 25.32% -> 25.28% (about -0.04 pp)
• First-choice share: 34.65% -> 31.67% (-2.98 pp)

So the huge horizontal movement in raw supporter counts is mostly a turnout effect. Koishi did not gain additional market share in 2026.

The character-specific change that remains after normalization is actually a decline in first-choice concentration.

That is strong evidence against the idea that the 2025 first-choice spike was simply continuing to strengthen year after year.

15. But the 2025 change did not completely disappear either​

If we compare only the endpoints - the 20th and 22nd polls - Koishi is still well above her 2024 baseline.

Two-year net change​

• First-choice votes: +1,828
• Non-first-choice supporters: +1,530
• Support rate: +2.52 pp
• First-choice share: +5.90 pp

This creates an important three-stage picture:

Shock -> partial reversion -> persistence

Click to expand...

or, more intuitively:

Spike -> broadening -> elevated new level

Click to expand...

The 22nd poll did not erase the 21st-poll change. It partially reversed the concentration while broader support expanded.

There is also a methodological lesson here: if we looked only at the 20th and 22nd endpoints, Koishi's growth would look much more balanced. The intermediate 21st-poll observation reveals that the two components did not rise together. First-choice support surged first; broader support caught up one year later.

Endpoint comparisons can hide path dependence.

16. What does the 22nd poll do to the competing explanations?​

The new data do not select one explanation, but they do change how convincing some stories are.

"Koishi's popularity itself is artificial"​

This becomes much harder to defend.

Koishi has a broad and persistent supporter base, strong demographic patterns, strong connections to Satori / Flandre / Hartmann / Subterranean Animism, and large non-first-choice growth in the 22nd poll.

The 22nd poll provides strong evidence that Koishi has a very large and broad measured support base.

"34.65% was simply a permanent new first-choice regime"​

This becomes weaker.

The first-choice share fell from 34.65% to 31.67% rather than remaining flat or strengthening.

"The 21st poll saw unusually strong legitimate campaigning"​

Still entirely plausible.

A coordinated campaign telling existing fans to rank Koishi first could produce a sharp first-choice spike without an equally sharp immediate increase in broader support.

"There was a real sudden preference shift toward making Koishi the favorite"​

Also plausible.

A fandom can experience a rapid favorite-conversion event before surrounding cultural indicators fully catch up.

"The 21st poll included prohibited duplicate or repeated-identity voting"​

The public aggregate data cannot rule this out either.

But they also cannot identify it.

We do not have public voter-level identity, device, IP, time-series, ballot-template, or duplicate-account data. That is the central limitation.

17. A statistical anomaly is not the same thing as an officially confirmed improper-vote case​

There is a useful historical control case.

In the 17th Touhou Project Popularity Poll, the organizing committee publicly stated that it had found several improper votes and that character rankings might change. The organizers subsequently continued the aggregation process, and the later detailed release contained revised totals and rankings. Contemporary archives preserve the official statement. [8][9]

That is a completely different evidence category.

17th poll​

Official confirmation + later revised results

21st poll​

Strong aggregate statistical anomaly + no official confirmation of improper voting

The existence of the former does not prove the latter.

But it is useful because it shows what a genuinely confirmed improper-vote case looks like: an organizer statement and an observable correction, not simply an unusual scatter plot.

18. Four things that should not be collapsed into the word "cheating"​

A lot of arguments about popularity polls become confused because several different behaviors are treated as if they were identical.

1. Ordinary campaigning​

"Please vote for Koishi."

Normal fandom behavior.

2. Incentivized campaigning​

"Vote for this character and show a screenshot for a chance to win merchandise."

Potentially controversial as a matter of community norms, but not automatically equivalent to duplicate voting.

3. Strategic ranking​

"Put Koishi first because first place is worth three points."

Potentially very important under the 3-2-1 system, but not automatically against the rules.

4. Duplicate / repeated-identity voting​

One person violating identity or voting restrictions to generate multiple ballots.

This is the category that requires actual audit evidence.

Treating all four as the same thing makes serious discussion almost impossible.

19. What the public data support - and what they do not​

I think the public data support these statements quite strongly:​

• Koishi's 2025 first-choice concentration was historically unusual.
• The 20th -> 21st change was unusual both in raw counts and after normalization.
• The shift was concentrated specifically in first-choice support rather than broad support.
• The 3-2-1 system made that concentration electorally important.
• The same first-choice explosion did not repeat in 2026.
• The 22nd poll instead showed partial mean reversion and broad non-first-choice growth.
• Koishi's measured support is broad, persistent, and embedded in a clearly observable fandom ecosystem.
• The 21st-poll spike therefore looks like a special event inside a real long-term popularity trend.

The public data do not establish:​

• how many first-choice voters selected only one character,
• the average number of characters selected by Koishi-first voters,
• the joint distribution of first and second choices among Koishi-first voters,
• repeated ballot-template patterns,
• temporal clustering of ballots,
• device / IP / account duplication,
• or the identity of any voter.

Without those data, several causal mechanisms remain observationally compatible with the same aggregate pattern.

Conclusion: an outlier, not a verdict​

The most interesting thing about Koishi Komeiji's 21st-poll victory is not simply that she won.

She was already one of Touhou's biggest characters, and the 22nd poll makes her genuine popularity even harder to dispute.

The unusual part is narrower and more specific.

Between the 20th and 21st polls, Koishi's support became suddenly and dramatically more concentrated in the first-choice position:

• first-choice votes: +1,799
• non-first-choice support: +77
• first-choice share: 25.77% -> 34.65%

That change was extreme relative to both her own previous trajectory and the contemporary Top-100 distribution.

One year later, however, the pattern did not repeat:

• first-choice votes: +29
• non-first-choice support: +1,453
• first-choice share: 34.65% -> 31.67%

The three-year sequence is therefore best described as:

A long-term popularity trend experienced an unusually sharp first-choice concentration shock in 2025, followed in 2026 by partial mean reversion, broader support, and persistence above the older baseline.

Click to expand...

That is an empirical description.

Its cause is a separate question.

The pattern is compatible with several mechanisms: a real preference shift, unusually strong but legitimate campaigning, strategic use of the ranking system, or other voting behavior. Public aggregate data cannot distinguish among them.

So the strongest conclusion I am comfortable making is not an accusation.

It is a request for better auditability - and more precise language when discussing unusual voting data.

An anomaly can be demonstrated. Its cause cannot.

Click to expand...

Methodology appendix​

A. Data reconstruction​

For the 20th poll, the older scoring system can be written as:

Score = N + F

so:

N = Score - F

For the 21st and 22nd polls, the 3-2-1 system gives:

Score = N + 2F + S

where S is the number of second-choice votes, so:

N = Score - 2F - S

Then:

Non-first-choice supporters = N - F

Support rate = N / total character voters

First-choice share = F / N

B. Cross-sectional outlier plots​

For the 20th -> 21st comparison, the main comparable Top-100 sample contains 99 characters after excluding a new character that had no previous-year observation.

For the 21st -> 22nd comparison, the comparable sample contains 96 characters.

Two spaces are used:

1. Raw counts: change in non-first-choice supporters x change in first-choice votes.
2. Normalized rates: change in support rate x change in first-choice share.

The source analysis uses Tukey IQR fences (1.5 x IQR and 3 x IQR) as descriptive outlier screens. Crossing an IQR fence means that a value is unusual relative to the contemporary distribution. It does not produce a probability of cheating.

C. Historical comparison​

A fixed panel of long-running top characters is used to compare annual first-choice-share changes before the 21st poll. The 96 prior changes are an empirical historical reference, not an iid statistical sample.

For that reason, z-scores, MAD-based scores and empirical tail counts are used as descriptive anomaly scales, not as literal causal probabilities.

D. Questionnaire limitations​

The demographic questionnaire is voluntary. Non-response may be non-random (MNAR), and the questionnaire describes Koishi supporters in general, not specifically Koishi-first voters.

It should therefore be treated as supporter-profile evidence, not as a direct explanation of the first-choice spike.

E. Co-voting limitations​

Published co-voting tables are aggregate, not full anonymous ballot records. They cannot tell us the complete ballot structure of Koishi-first voters.

Some published pairwise common-voter counts are also internally inconsistent in opposite directions (A->B vs B->A). Robust analysis should treat those discrepancies as intervals or run sensitivity checks instead of silently choosing one side.

F. What would actually resolve the causal question?​

The most useful additional data would be anonymized audit summaries such as:

• distribution of ballot size among first-choice voters,
• first-choice x second-choice joint tables,
• single-selection ballot rates,
• repeated ballot-template frequency,
• coarse time clustering,
• number of ballots removed by anti-duplicate checks,
• and a public description of invalid-vote / duplicate-vote handling.

These would improve auditability without requiring the organizers to reveal personally identifying voter information.

Public references​

[1] 20th Touhou Project Popularity Poll - official results
https://toho-vote.info/results/20

[2] 21st Touhou Project Popularity Poll - official results archive
https://toho-vote.info/results/21

[3] 22nd Touhou Project Popularity Poll - official result portal
https://toho-vote.info/result

[4] THBWiki - 20th poll, character division
https://thwiki.cc/东方系列人气投票/第二十回/角色部门

[5] THBWiki - 21st poll, character division
https://thwiki.cc/东方系列人气投票/第二十一回/角色部门

[6] Cross-check dataset for the 19th-21st polls (GitHub)
https://github.com/xiangze/toubou_vote_analysis/tree/main/data/19_21

[7] Chinese fan analysis of the 21st poll, including co-voting relationships
https://www.bilibili.com/opus/1241248116965900295

[8] Contemporary archive quoting the official @toho_vote notice of Oct. 5, 2021 that several improper votes had been found
https://danmakublog.com/archives/10929799.html

[9] Contemporary follow-up archive recording the later detailed / revised release for the 17th poll
https://danmakublog.com/archives/10939862.html

I guess that's what mathematicians are left to do after graduating

Holy shit, that's a demonstration of strength right there. I'm seriously impressed. I hope you're doing this to flex or practice your skills.

I'll give it a more profound read when I got time. Thing is, from the demographic (in percentages) and the sudden spike: there's obviously got to be a "where" exactly. Thing is, I very much like forensics. With your post, there's a good starter idea of what kind of person is the (average) Koishi voter. I want to know what made Koishi popular in the first place and why, and the other characters tied to her.

I don't know how to put it correctly but I'm just doing this because I want to. It's really more to get to the source of it all. Still, impressed at the amount of work this is. Catch you later, got many more limbs in pots.

kanedafanatic said:

Holy shit, that's a demonstration of strength right there. I'm seriously impressed. I hope you're doing this to flex or practice your skills.

I'll give it a more profound read when I got time. Thing is, from the demographic (in percentages) and the sudden spike: there's obviously got to be a "where" exactly. Thing is, I very much like forensics. With your post, there's a good starter idea of what kind of person is the (average) Koishi voter. I want to know what made Koishi popular in the first place and why, and the other characters tied to her.

I don't know how to put it correctly but I'm just doing this because I want to. It's really more to get to the source of it all. Still, impressed at the amount of work this is. Catch you later, got many more limbs in pots.

Click to expand...

I think one of the reasons Koishi became popular is because of her more "enigmatic" character and the fact that even in-lore, it's stated something like she's "an imaginary childhood friend you used to play with". Her inability to feel emotions right and how she disappears and is forgotten by others' consciousness is I think another aspect of what makes her popular.
Plus, her design probably grabs a lot of attention as well.
Personally, despite Koishi being an interesting character (and to some extent or another, I empathize with her), I find the other members of the Komeiji household to be more interesting to me. (Well, for now at least.)

This was quite an interesting read. I would like to point out that regular changes in voting systems makes data comparision less. Data from the 22nd vote can be only compared to data from the 21st vote, but not with the 20th vote. For example there is the second place vote which can affect the data too. It is interesting that this year this group increased for Koishi, but not the first-place voters. Of course it doesn't explain the anomaly with first-place voters for Koishi, but second-place voters could affect other characters data.
It is quite sad that there is no demographics for first-place voters. But I think young fans could give a lot of first-place votes for Koishi (it is impossible to check it). And there are quite a lot of new young fans in Touhou.
Anyway, looking at the data it looks like there are some other interesting throws this year. It could be interesting to analyze them too.

SateriCat17 said:

I think one of the reasons Koishi became popular is because of her more "enigmatic" character and the fact that even in-lore, it's stated something like she's "an imaginary childhood friend you used to play with". Her inability to feel emotions right and how she disappears and is forgotten by others' consciousness is I think another aspect of what makes her popular.
Plus, her design probably grabs a lot of attention as well.
Personally, despite Koishi being an interesting character (and to some extent or another, I empathize with her), I find the other members of the Komeiji household to be more interesting to me. (Well, for now at least.)

Click to expand...

I see what you mean. Rather, what I meant was what are the factors in the rise of Koishi's popularity in the last two years? That she is a fascinating character or not, that isn't what I'm looking for. At least, it's a factor I consider but not the biggest one. Basing off Point 8, about the age demographic: it happens as well that a lot of fans voted for their first time and choose Koishi. Not only that, the main majority of the voters are situated in the 10-14 and 15-19 range, in the sample. Age matters because there's a plausible reasoning to assume that those younger fans have been introduced to their first Touhou-related piece of content where Koishi appears. What I'm looking for is what exactly introduced those voters to Koishi and Touhou by extension. Why her in particular? Which piece(s) of content have the majority watched or interacted with so that Koishi's popularity booms for the last two years?

It's a bit ridiculous but, that's what fascinates me here. The foundations to this whole thing. It could've been anybody as far as I'm concerned to take the 1st role. If Koishi has been on the rise since those two years, then what supports it? I get the point that she is recognisable by her design and that she's enigmatic as a character, but that isn't enough. There's got to be something that explains it from an outside means (media content) I do have some theories in my mind but those are unproven yet.

vanyk said:

This was quite an interesting read. I would like to point out that regular changes in voting systems makes data comparision less. Data from the 22nd vote can be only compared to data from the 21st vote, but not with the 20th vote. For example there is the second place vote which can affect the data too. It is interesting that this year this group increased for Koishi, but not the first-place voters. Of course it doesn't explain the anomaly with first-place voters for Koishi, but second-place voters could affect other characters data.
It is quite sad that there is no demographics for first-place voters. But I think young fans could give a lot of first-place votes for Koishi (it is impossible to check it). And there are quite a lot of new young fans in Touhou.
Anyway, looking at the data it looks like there are some other interesting throws this year. It could be interesting to analyze them too.

Click to expand...

Yeah, I think your point about the rule changes is important. The introduction of an explicit second-choice vote means that the behavioral environment is not identical across the 20th, 21st and 22nd polls, even if some quantities such as total supporters, first-choice voters and first-choice share can still be reconstructed consistently.

So I probably should be more careful in how I describe the 20th-to-22nd comparison. I would treat it as a descriptive endpoint comparison rather than a perfectly rule-invariant comparison. The 21st-to-22nd comparison is definitely cleaner because the scoring system is the same.

I also agree that the increase in Koishi's second-choice voters in the 22nd poll is interesting. Her first-choice count barely changed, while second-choice and other non-first-choice support increased substantially. That is part of why I describe the 22nd result as "broadening" rather than another first-choice spike.

As for the demographics, I think we have exactly the same problem in mind. Koishi supporters are younger and more likely to be first-time participants, but unfortunately we do not have demographics specifically for Koishi-first voters. So the idea that younger/newer fans contributed heavily to the first-choice spike is plausible, but currently untestable with public data.

And yes, I also noticed that there are several other interesting movements in the 22nd poll. I think comparing those characters systematically would be a useful next step, especially as a contemporary control for Koishi.

kanedafanatic said:

Holy shit, that's a demonstration of strength right there. I'm seriously impressed. I hope you're doing this to flex or practice your skills.

I'll give it a more profound read when I got time. Thing is, from the demographic (in percentages) and the sudden spike: there's obviously got to be a "where" exactly. Thing is, I very much like forensics. With your post, there's a good starter idea of what kind of person is the (average) Koishi voter. I want to know what made Koishi popular in the first place and why, and the other characters tied to her.

I don't know how to put it correctly but I'm just doing this because I want to. It's really more to get to the source of it all. Still, impressed at the amount of work this is. Catch you later, got many more limbs in pots.

Click to expand...

Yes, I think this is exactly the more interesting question, and I am very happy to see this.

"Why is Koishi an appealing character?" and "Why did Koishi gain so much support among younger/newer fans in this particular period?" are not really the same question.

Her design, personality, themes of unconsciousness, isolation, being forgotten, etc. can help explain why people become attached to her after encountering her. But those traits have existed for many years. By themselves, they do not explain why the growth became especially visible recently.

What I would really like to know is the exposure pathway: what Touhou-related content introduced younger fans to Koishi in the first place? Was it a game, LostWord, fan animation, music/remixes, memes, YouTube/Shorts, TikTok-style content, fanart, Fumo culture, or something else?

The demographic data make this especially interesting because Koishi supporters are younger than the overall sample and include more first-time poll participants. But unfortunately, we do not know the demographics specifically of Koishi-first voters, so there is still a missing link between "young/new fans" and the first-choice spike.

You mentioned that you already have some theories about the media side of this. I'd actually be very interested in hearing them, even if they are still unproven. If there is a specific piece of content or platform whose popularity rose at roughly the same time, that could give us a much better starting point for investigating the mechanism.

Alright, I decided to look for Japanese researches on this topic. Well, there is no clear explanation of this phenomenon by them too.
Looking at statistics if we will consider that all votes have the same weight then the situation wouldn't change much for top 15 characters (the only differences are Koishi, Marisa and Reimu taking 3rd, 1st and 2nd places; Yuyuko and Alice - 13th and 12th). Using old voting system (2 point for 1st place, 1 point for other places) didn't change the situation much too:

For less popular characters this consideration would change their places more seriously. Anyway researching second place votes only showed that more people second-place voted for percent for less popular characters:

But nothing else interesting.

By the way, I found an article in which comments for Koishi, Reimu and Marisa were analyzed. They found out that there were more Japanese comments for Marisa while there were more non-Japanese comments for Reimu and Koishi. In addition there were much more Korean comments for Koishi. Another interesting note is that the more Japanese voted for the character the more comments it had (the only exception is Cirno, because of ⑨ comments), so it could be important to consider that too. Still I doubt this can help to fully explain the anomaly.

It is important to note that there quite a lot of young Touhou fans nowadays. Anyway most Japanese fans agree that this demographic group is underresearched and isn't fully understood. It is important to understand that old Touhou fans are quite a minority now. There were some discussions about this topic, but there was no decisive conclusion.

不人气的大狸子 said:

What I would really like to know is the exposure pathway: what Touhou-related content introduced younger fans to Koishi in the first place? Was it a game, LostWord, fan animation, music/remixes, memes, YouTube/Shorts, TikTok-style content, fanart, Fumo culture, or something else?

Click to expand...

That is quite a good question.
1. Original games. I really doubt this, because you need a computer to play them, and don't many Japanese school students use computers.
2. LostWord. It could boost Koishi's popularity, but looking at official Japanese YouTube channel and Twitter account videos and posts with Koishi are not the most popular ones (but still quite popular).
3. Fan animations. Hmm, not sure about this, but there is this popular animation by Korean fan and there are a lot of Japanese comments. I
4. Music/remixes. It could be. Though it is interesting that according to this database there are around the same amount of arrangements of Koishi and Satori themes.
5. Memes. Perhaps? It is difficult to check this though.
6. YouTube/Shorts. I suppose YouTube, Pixiv and Twitter are the main sites young Japanese Touhou fans can use.
7. TikTok. Hmm, I think this app could be popular among younger fans, but I don't know much about it.
8. Fanart. I think it could affect too.
9. Fumo culture. Perhaps? Not sure how to check it.
Some my ideas:
1. Yukkuri commentary. I thought they could affect the situation, but Marisa, Reimu and Youmu appear much more frequently in such videos. So I doubt this.
2. Non-Fumo Merchandise. Koishi related merchandise appears frequently and I see her on posters related to commercial collabs, but this doesn't explain why Koishi was added there in the first place and why she is that popular among younger fans as other Touhou characters appear more frequently.

Thinking about it there could be one more question: "Why Koishi is more popular than Flandre even though they have a lot in common and there is more Flandre related content". To be honest, I don't know why.

Edit: I have no idea why this site corrupts inserted tables, so I replaced them with screenshots.

Okay. What I do right now is a draft where I show the ways to get the answers than the theories. Somewhere to begin; thinking it out loud and reframing it. As such, don't take any of it at face value unless when pointed out.

Now, where @不人气的大狸子 (nicknamed OP from there) analyzed about how and what exactly made Koishi rise to the first of this year's poll: I appeared and pondered how did Koishi rose to the top like this? "What Touhou-related content introduced younger fans to Koishi in the first place?"; formulated by the OP. That is the core question meant to drive my reverse-engineering investigation towards a plausible conclusion (forensic wasn't the right word: soz.) I will also base upon OP's analysis and this thread's answers to lay down my ideas.

So, what I will concentrate first is who and what. Who is the kind of person that voted for Koishi, first pick or supporter (N and F in OP's analysis), and what ties them together?

To begin the first part: I will say that OP's analysis gives an idea of what ties Koishi voters together. Basing off Point 8 and Point 10 respectively, OP explains the age demographic of those who vote in the overall sample and in particular, what did they also vote along with Koishi

> 67.37% (10-14 and 15-19 combined) of Koishi voters in the 21st poll were between pre-teens and young adults from the overall sample. In other words, like OP pointed out, more than the half of the voters are young fans and the fanbase is young. (Point 8)
> Koishi voters from the 21st poll are also very much tied to the terms used in Point 10; Satori Komeiji, Flandre Scarlet, songs like Hartmann's Yôkai Girl and Last Remote and the game Subterranean Animism. What this points at is that there are links to those subjects; and Flandre stands out of the bunch.

However, this is insufficient. It doesn't paint a clear picture of our crowd. It gives clues as to how they are and where to look but, this is barely nothing in actual usage.

This is why I thought to reframe my idea: because it lacks a clear idea of who we are talking about. What is the link between Koishi and Flandre, as @vanyk pointed out earlier? What do they have in common? If we don't know who we are talking about first, then we don't know what ties Koishi fans together and the why to the abnormal rise of Koishi's first-vote popularity.

vanyk said:

Looking at statistics if we will consider that all votes have the same weight then the situation wouldn't change much for top 15 characters (the only differences are Koishi, Marisa and Reimu taking 3rd, 1st and 2nd places; Yuyuko and Alice - 13th and 12th). Using old voting system (2 point for 1st place, 1 point for other places) didn't change the situation much too:

Click to expand...

I have always felt that the Japanese side tends to lag behind a little when it comes to this kind of analysis. At least in the Chinese Touhou community, someone had already worked out the unweighted ranking on the very day the poll results were released. And among the articles I have been able to find so far, none of them seem to go into as much detail as mine.

vanyk said:

By the way, I found an article in which comments for Koishi, Reimu and Marisa were analyzed. They found out that there were more Japanese comments for Marisa while there were more non-Japanese comments for Reimu and Koishi. In addition there were much more Korean comments for Koishi. Another interesting note is that the more Japanese voted for the character the more comments it had (the only exception is Cirno, because of ⑨ comments), so it could be important to consider that too. Still I doubt this can help to fully explain the anomaly.

Click to expand...

In fact, I also found that article analyzing the languages used in first-choice comments. If I remember correctly, it was written by a Korean author. I did not pay much attention to it at first, because it does not really have any decisive bearing on the conclusions of my article. However, someone on Baidu Tieba once suggested that perhaps some voters were confusing Koishi with Satori, so I did use that article as one example to argue against that idea.

vanyk said:

It is important to note that there quite a lot of young Touhou fans nowadays. Anyway most Japanese fans agree that this demographic group is underresearched and isn't fully understood. It is important to understand that old Touhou fans are quite a minority now. There were some discussions about this topic, but there was no decisive conclusion.

Click to expand...

This is also similar to what I mentioned before: the increasingly young age profile of the fandom is something almost every Touhou community seems to be facing. But it is also quite obvious that this younger demographic has not yet gained much influence or visibility within the community itself.
As a side note, I cannot help but find Reimu's performance in the popularity poll rather remarkable. Her voter base seems to be gradually aging, and her first-choice rate is not particularly high among the top five characters, yet she somehow continues to remain firmly in the top three.




Unfortunately, the Chinese Touhou community has already fallen into a fairly intense flame war over this result, and some of the arguments have even escalated into attacks on Koishi herself as a character. One of the main targets of criticism is Koishi Komeiji's Heart-Throbbing Adventure, a highly controversial fan work which, according to these critics, attracted a large number of "bad" fans to Koishi.

There is also a broader complaint that Koishi's personality has been heavily distorted in fan interpretations. She is often portrayed as someone who is depressed, bullied, socially isolated, or emotionally damaged—I am honestly not quite sure what the best English term would be here—and some people therefore claim that she appeals disproportionately to psychologically troubled or socially dysfunctional people who identify with that kind of image.
I should stress that I am describing arguments I have seen in the Chinese community here, not endorsing them myself, and I hope this wording does not offend Koishi fans.

不人气的大狸子 said:

There is also a broader complaint that Koishi's personality has been heavily distorted in fan interpretations. She is often portrayed as someone who is depressed, bullied, socially isolated, or emotionally damaged—I am honestly not quite sure what the best English term would be here—and some people therefore claim that she appeals disproportionately to psychologically troubled or socially dysfunctional people who identify with that kind of image.

Click to expand...

Now, I'm not the biggest Koishi fan, far from it, but I think that's pretty much a fact. KKHTA is probably the biggest example of it, as old and corny as it is. Maybe not the biggest factor of them all to the re-interpretations but, very likely the source of it all.

不人气的大狸子 said:

I have always felt that the Japanese side tends to lag behind a little when it comes to this kind of analysis. At least in the Chinese Touhou community, someone had already worked out the unweighted ranking on the very day the poll results were released. And among the articles I have been able to find so far, none of them seem to go into as much detail as mine.

Click to expand...

The problem is that most of analysis by Japanese is posted on Twitter, so it may be difficult to find it. It can be useful to use Nitter to find such research posts or follow Japanese Touhou researchers. By using Nitter I found these threads[1][2][3][4] and posts by _beslin quickly (if I remember correctly I found some more posts later but they didn't help a lot). So such things exist, you just need to know where to find them. (Well, it looks like that is how things work in the East).
I agree that your article is the most in-depth research on the topic of Koishi popularity on Touhou Vote. That you for your research.

不人气的大狸子 said:

In fact, I also found that article analyzing the languages used in first-choice comments. If I remember correctly, it was written by a Korean author. I did not pay much attention to it at first, because it does not really have any decisive bearing on the conclusions of my article. However, someone on Baidu Tieba once suggested that perhaps some voters were confusing Koishi with Satori, so I did use that article as one example to argue against that idea.

Click to expand...

Hmm, there was a similar article by a Korean author? It would be interesting to read it too.
The one I sent was by Japanese author and all comments were analyzed. I agree that it doesn't solves the anomaly, but it is still interesting that there were quite a lot of Korean comments (considering that characters more popular outside of Japan have smaller comment rate, so it is difficult to say how many Korean votes are there). And I really doubt confusing Koishi and Satori could be a reason.

不人气的大狸子 said:

But it is also quite obvious that this younger demographic has not yet gained much influence or visibility within the community itself.

Click to expand...

I think the main reasons for that is younger fans using other platforms (for example TikTok) and producing less fanworks (well, they don't have much experience and it may more difficult for them to do big projects in general).

不人气的大狸子 said:

As a side note, I cannot help but find Reimu's performance in the popularity poll rather remarkable. Her voter base seems to be gradually aging, and her first-choice rate is not particularly high among the top five characters, yet she somehow continues to remain firmly in the top three.

Click to expand...

Somewhat agree. She doesn't have a lot of first-place votes, but she has quite a lot of votes in general (she is on the third place by the number of votes), but I don't think that is really surprising.

不人气的大狸子 said:

Unfortunately, the Chinese Touhou community has already fallen into a fairly intense flame war over this result, and some of the arguments have even escalated into attacks on Koishi herself as a character.

Click to expand...

That is sad to hear that. That is just a popular vote, it doesn't affect anything to be honest.
I really doubt KKHTA is the main reason for her popularity and that most of young fans watched it.
That is quite a common problem among Touhou characters. Their fanon interpretations may be very different from their canon versions. I think she may attract depressed people but that doesn't explain her popularity among young fans.

I think her popularity is a combination of different factors. Some factors could affect her popularity among school students the others could affect her popularity among university students. But this requires more research.

vanyk said:

I really doubt KKHTA is the main reason for her popularity and that most of young fans watched it.

Click to expand...

that's not what I meant. Rather that the current re-interpretations of Koishi in the fandom like OP pointed out stands mainly from there. You're right that it has no influence in the current subject. I directly answered the question directly as it's own thing.

kanedafanatic said:

that's not what I meant. Rather that the current re-interpretations of Koishi in the fandom like OP pointed out stands mainly from there. You're right that it has no influence in the current subject. I directly answered the question directly as it's own thing.

Click to expand...

Hmm, I see. These fan animations could make this interpretation more popular. But I think TH14.5 could also boost this interpretation. It is difficult to check what was the main reason for this. (perhaps Pixiv tag analysis by year might help? I don't know).

vanyk said:

Hmm, I see. These fan animations could make this interpretation more popular. But I think TH14.5 could also boost this interpretation. It is difficult to check what was the main reason for this. (perhaps Pixiv tag analysis by year might help? I don't know).

Click to expand...

I'll detour away from the main subject for this one and focus on that. It'll give me time to continue reframing my idea earlier on, too.

I didn't pay attention (or just didn't care in the first place) about Touhou 14.5 so, I'm as clueless as you are as to how it boosted Koishi's "broken" fan-interpretation. KKHTA is an easy pick given the tone the series have and the re-interpretations' centered about having a "misunderstood, broken character" of some kind. As to which recent piece of content? Fuck if I know.

Actually, maybe @Ken Hisuag could help with music-related content. Both between the time period of the 20th and 22nd polls and this random subject we're having here. Having his expertise on the matter could really be of use as to determine if any Koishi-related music has blew up over on social media or so.

Hell, now that I think of it: as much as I don't use TikTok, what the fuck is even happening there, Touhou-wise? Barely gets talked about but maybe there's an answer here? I'm generalizing like an old man here but, given the young fanbase and the kind of content Koishi is involved in, there could be our answer here.

kanedafanatic said:

Hell, now that I think of it: as much as I don't use TikTok, what the fuck is even happening there, Touhou-wise? Barely gets talked about but maybe there's an answer here? I'm generalizing like an old man here but, given the young fanbase and the kind of content Koishi is involved in, there could be our answer here.

Click to expand...

Hmm, I think we found a serious problem here. We don't know much about younger Touhou fans. We don't know which platforms they use the most and what kind of content they watch more. In Japan case it is somewhat easier because Twitter and Pixiv are still the main platforms for artworks. But we cannot be sure about video content. Touhou is not popular on Niconico Douga anymore, Touhou content is more popular on YouTube, but I am not sure this is the main platform among younger fans. I think TikTok or TikTok-like app might be more popular. It requires a serious research.
(Another thing that is popular among Touhou fans is Touhou doujin sphere, but analyzing it might be very difficult as we don't know about all fanworks from there (this thing is not documented))

Goes in full circle about who is the Koishi fan and what all over again... Well, no shit I said reverse-engineering is my kind of tea. Well then, give me some time so I can flesh out my idea and some path and I'll share it here.

One thing I find interesting is that Chinese Touhou fans often bring up Touhou Hero of Ice Fairy as a possible factor, especially its Koishi-focused DLC Rose Idol, released in April 2025. Strangely, none of the replies here have mentioned it so far. That may itself show how geographically fragmented Touhou media exposure has become: something that feels highly visible in the Chinese community may barely enter the explanatory map of Western fans.

不人气的大狸子 said:

One thing I find interesting is that Chinese Touhou fans often bring up Touhou Hero of Ice Fairy as a possible factor, especially its Koishi-focused DLC Rose Idol, released in April 2025. Strangely, none of the replies here have mentioned it so far. That may itself show how geographically fragmented Touhou media exposure has become: something that feels highly visible in the Chinese community may barely enter the explanatory map of Western fans.

Click to expand...

Interesting. I was targeting more on the Western side of things with the way that the thread was headed. Yet, maybe I'm reading it wrong, isn't it like the other Asian parts of the Touhou community not as affected, like with @vanyk's analysis of Touhou on video-sharing websites?

Given how fragmented the Touhou media exposure is, as you put it, maybe it just didn't catch on as much as here. There I was, looking in the direction of TikTok as I'm pretty sure a majority of Touhou fans turn a blind eye on that part of the fanbase. Reddit's probably worth taking a look too, even if I believe that this whole website is as interesting as a septic tank.

Should probably haul ass a bit to get that who and what thing going on. Feels like this investigation is stalling because of that (I.E., because of me.) Then again, probably I'm putting too much importance on it, and I sound like a pretentious prick while at it.