Song Lists vs Real Games: What 33,770 Picks Say

Two things shipped on October 2 and 3 that didn't look related. One reworked the song lists. The other changed four numbers on the home page. They met while I was writing this up, because the counters that fixed the home page are also the only lasting record of what people pick in a game. And that's the honest way to check whether a list of "best hype songs" is any good: see whether anyone plays them.

Every number below was pulled on October 6, 2026. Song and rating data counts from August 21, the day the stats tables started.

What songs do players actually pick?

Mostly the ones you'd guess. Of the songs that also appear on one of our lists, Billie Jean has been picked 50 times, Eye of the Tiger 45 and Mr. Brightside 40. That counts every version of a track (single edit, remaster, live) as one song. Across all 19,500 different tracks anyone has submitted, 14,627 were picked exactly once, so three in four songs only ever show up in one round.

Getting picked and scoring well are different skills. The average rating for any song in any round is 3.08. Thunderstruck gets picked all the time and lands at 2.78. My read is that it's the first song half the room thinks of, so it wins on recognition and loses on surprise. Eye of the Tiger is the opposite. It's just as obvious and it still scores 3.46, across 126 ratings, more than any other song in that table.

Which songs score best and worst?

Among list songs with at least 20 ratings, Bangarang by Skrillex scores highest at 4.38 out of 5, and Shake It Off by Taylor Swift scores lowest at 2.02. Twenty ratings is my cutoff because below that, one loud room can move a song by a whole star.

Fast Car at 2.53 hurts me personally. Let It Happen is the one I'd actually argue with: 27 picks, so people clearly love it, and a 2.35 average. I have a theory I can't prove from these numbers. It runs 7 minutes 47 with a long build, and nobody in a battle waits for a build.

How do the song lists hold up against real games?

About half of each list gets played. Of the 801 different songs across the 26 lists, 434 (54%) have been picked at least once, and together they account for 3,042 picks. The Songs to Scream in the Car list is the closest thing to a perfect match: 45 of its 50 songs have turned up in a real game.

The loud lists get played and the quiet lists get rated. The five most-picked lists all average between 3.19 and 3.34. Dinner party, yacht rock and Fleetwood Mac each have fewer than 200 ratings, and all three average above 3.5. I think that's selection more than quality. Nobody brings Steely Dan to a battle by accident. You bring it when the prompt is begging for it, and the room rewards the bit.

It's also a useful check on what I just did. Four of the lists I grew to 50 tracks on October 2 (dinner party, fall asleep, yacht rock and 90s R&B) are among the least played in games. They're lists for reading, or for putting on while you cook, and that's a fine thing to be. I just shouldn't expect them to send many people into a room.

What changed on the song lists?

Seven lists grew from 15 tracks to 50, the hype list got a new title, and every list got a button that starts a game on its own prompt. Here's the full set:

• Scream in the Car, Sad Songs That Hit Hard, Songs to Cry To, Fall Asleep To, 90s R&B, Yacht Rock and Dinner Party each went from 15 tracks to 50. Across the 26 lists that's 914 tracks, up from 634, and every one still has its own line on why it's there.
• The hype list is now titled "Best Hype Songs of All Time", with "Hit in 4 Seconds, Clean Flagged" after it.
• "Start a room on this prompt" sits under the top pick and again after track 10. The room it opens is already on the list's theme, with the list's own prompt first in round 1.
• Lists longer than 20 tracks get an "If you liked this" row after track 20, pointing at the next list.
• On the aux battle categories page, the spinner now spins right there instead of sending you to the wheel page, and "Play this category" starts a room on whatever it landed on.
• A Spotify playlist for each list is built but hidden. The code is in; the playlists themselves don't exist yet.

Why rewrite the title and not the description?

Because Google wasn't showing the description. For a list page it builds the snippet out of the H1 and the first few track names, so our result read "Best Hype Songs · Eminem - Till I Collapse · Kanye West - POWER" no matter what I wrote in the meta tag. The title and the H1 are the only words I actually control in that result.

The bigger finding was less fun. Search Console said the hype list averaged position 5.99, with 11,731 impressions and a 0.81% click rate over 28 days. When I searched "hype songs" myself, it was result 21, top of page 3. Search Console only counts an impression when the page gets shown, so the average was built from the rare searches where we made page 1. The real problem is rank. A title fixes the cheap half.

Why put a game button in the middle of a list?

Because people were reading the whole list and then leaving. In the 30 days before the change, the hype list got 117 visits from Google. They read for a median of 70 seconds and scrolled about halfway, which is a reader who's getting what they came for. Not one of them started a room. Nothing on the page told them there was a game attached.

The categories page had the opposite problem. About 37% of visitors from Google pressed "Spin for a Category", and that button took them to the wheel page, where 2.9% start a room, compared with 8% of people who stay on the categories page. The most popular button on the page was sending people to the page where they're least likely to start a game. So the spinner moved onto the categories page itself.

Has it worked?

Too early to tell, and I'm trying hard not to pretend otherwise. From the deploy late on October 2 to today, the categories page had 158 sessions, those sessions spun the new spinner 430 times, and 9 rooms were started from "Play this category". The song lists had 155 sessions and nobody has pressed the new button yet. That zero goes in the log like every other number.

Two things make any early read shaky. Google's September spam update has been rolling out since September 24 and still hasn't finished, and the hype page took four changes at once. If its rank moves, I'll only be able to credit the batch, not any one change. The first proper read is October 23, 21 days after the deploy.

Why did the numbers on the home page change?

Because the old ones were counted the wrong way for a game that deletes its own data. The tiles said "Rooms played", "Players" and "Songs rated since August", and they were worked out by counting rows in the game's tables at build time. Room data is deleted 24 hours after a room is created, which is what the privacy page promises, so counting those rows measures about a day of play. It can't give an all-time total, and it can go down between two builds.

A number that can shrink isn't a counter. So I changed how the numbers are counted, and renamed the tiles so each one says exactly what it measures.

How do you count games after the data is deleted?

With a counter that goes up as things happen and never looks back at the rows. It is one database row with four numbers, and a trigger adds one each time something new arrives:

• A game counts when a room leaves the lobby. A room that never starts is someone clicking a button, not a game.
• A song counts when it gets submitted.
• A vote counts once. Changing your rating is an update, so it isn't counted twice. A Twitch chat tally is saved as one row however many people voted in chat, so this number runs low.
• A player counts each time they take a seat. One person in three games is three.

The trigger swallows its own errors. If it ever fails, the counter misses one. It can't stop a game from saving a song, which was the one thing I wasn't willing to trade for a nicer number. Songs and votes carry over from the stats tables, which started on August 21, so both are lower bounds.

Why "players joined" and not players?

Because I can't count distinct people without keeping who they are, and I don't keep that. Counting sign-ins doesn't work either. The site gives an anonymous session to anyone who clicks anything, and plenty of visitors never sit in a room, so that would count visitors, not players.

So the tile now says Players joined, currently 15,057, and it means seats. "Songs rated" became Votes, 76,896, with the lower-bound caveat above. The four numbers are the same everywhere: the home page (which refreshes them every minute and never shows less than the build did), the About page, the Aux Wars page, and the structured data search engines read. The stats tables keep totals per song and per prompt and nothing that points back at a person or a room. The privacy page says the same.

What I'm watching next

October 23 is the first real read on the song list changes. After that come the Spotify playlists, which I still have to create on the account.

The one I'm most curious about is a box on each prompt showing what people actually submit to it. Since September 30 the game has logged 4,033 prompt-and-song pairs, and only 9 of them have come up three or more times, with a top of 4. That's too thin to show without making something up, so it waits. If you want to speed it up, start a battle and pick something nobody else would.

How this was made: song numbers come from the game's aggregate tables, matched to our lists by title and artist with every version of a track counted once. Ratings are what rooms gave during play, 0.5 to 5 stars. Traffic numbers are from PostHog and Search Console. Written with AI help on the wording; the data pulls and the opinions are mine.