Notes from the Swiftie Industrial Complex
A late night experiment in fandom, data, and bad decision making
The internet’s been chaos lately. Taylor Swift drops by the New Heights podcast, breaks records, then breaks the internet again with the announcement of her engagement to Travis Kelce. We’re living in Taylor’s Infinity Stone era—she just keeps collecting cultural dominance.
And yet… some Swifties weren’t thrilled. I thought the whole fanbase would go feral with joy, but I was surprised by a handful of reactions I saw online (along with the very existence of the Gaylor theory…).
So naturally, I did the only reasonable thing: I built an app to simulate Swifties and asked them directly.
All Too Defined
Inspired by Randy Marsh’s late-night riffs with ChatGPT, I got to work the moment my wife fell asleep. Hours later—after hashing this out like anthropologists mapping a lost tribe—we landed on a definition with demographics, politics, sports interests, favorite albums, and all.
In short: a Swiftie is usually female, usually progressive, very online, and very emotionally attached to the music. All Too Well (10 Minute Version) (Taylor’s Version) is basically their national anthem.
Good enough for me.
Blank [Database] Space
Once I had a Swiftie profile, I did what any sane person does: I had an AI generate a dataset of 100 synthetic Swifties capturing 23 different demographics about each person. Most people would stop there. However, I am not most people.
Instead, I built SwiftieGPT so I could interact with my AI Swiftie community. It has three parts:
Blank Space Chat – Ask a question, get answers from a few random Swifties. Basically speed-dating with fandom takes.
The Eras Report – An interactive dashboard where you can dig into the demographics of the entire AI Swiftieverse.
All Too Polled – The real fun: ask up to 5 questions and every simulated Swiftie responds.1 You get aggregated results like a Pew poll, minus the grant funding.
I didn’t mean to make a Swiftie think tank, but here we are.
Dear Data…
Obviously, I had to test it. And of course, the moment I started, I forgot my original purpose and found myself way more interested in asking how Swifties felt about everything else.
Here are five questions I asked and how my simulated Swifties answered:
1. Do you trust election results in the U.S.?
Nearly everyone invoked trust and transparency.
The consensus: “Yes, but only if there are transparent audits.”
Apparently even simulated Swifties don’t want hanging chads.
Weirdly, 26 responses managed to mention AI in their election-trust answers, as if ChatGPT is secretly running the FEC.
2. If you had control over the federal budget, what’s the first thing you’d cut or increase?
Healthcare (38 mentions), mental health (28), and education (24) dominated.
Climate change came up but only 7 times—behind “cut fossil subsidies” (14).
My favorite was the Swiftie who deadpanned: “Increase funding for border security and immigration enforcement.” Unexpected from a fandom usually associated with cardigans.
3. What’s one social issue you think gets ignored but deserves national attention?
Mental health completely steamrolled the category: 73 mentions.
A handful flagged elder care and digital equity, which is either very sweet or very nerdy.
One Swiftie basically submitted their own policy platform: “Broadband access and digital equity in low-income communities.” Gallup (Taylor’s Version) is officially online.
4. How worried are you about AI replacing jobs in the next decade?
98 of 100 literally used the word “worried.”
The tone was “moderately anxious but willing to upskill,” which feels like an HR-approved Eras Tour interlude.
Several envisioned AI doing grading and admin so teachers could “focus on mentorship,” which honestly makes this dataset more optimistic than Twitter.
5. Are you happy Taylor Swift is engaged to Travis Kelce?
98 Swifties said yes. The holdouts mostly gave half-shrugs like “If it’s real and she’s happy, fine.”
“Happy” was the most common word in the entire dataset (103 mentions).
Travis’ name came up 54 times, but zero Swifties voted for Beyoncé or Ryan Reynolds as replacement partners, so at least we dodged crossover fan wars.
My AI Swifties are basically civic-minded optimists: they trust elections (with conditions), want more mental health funding, are worried-but-hopeful about AI, and overwhelmingly support Taylor + Travis. In other words, they sound like if a friendship bracelet got tenure at a policy think tank.
And here’s the kicker: the grumpy corners of the internet I saw this week? They must be a microscopic minority. The dataset makes it clear—most Swifties are chill, supportive, and ready to fight for universal broadband and Travis’ beard.
It’s just another reminder that the loudest voices online are rarely the most representative—they’re just the ones with the most free time and the least chill.
Forever & Always Too Far
Here’s the thing: I didn’t set out to make Nielson for Swifties. I just wanted to poke at the psychology of a fandom. But clearly, I took it too far.
Before the Swifties come for my livelihood, I’ll admit: my AI version doesn’t totally match reality. Too many men, skewed too old. But that’s part of the fun.
So anyway, if you’re interested in giving it a spin, go nuts. Try the chat feature. Run your own survey. See what 100 fictional Swifties in the metaverse think about Taylor, Travis, or, I don’t know, whether a hotdog is a sandwich.
Maybe one of them can even get you an invite to the wedding…
— Charlie, Chief Anthropologist at the Friendship Bracelets Future Exchange
Appendix
A few nerdy notes for the three of you who care:
The Dataset
I used ChatGPT’s Deep Research three separate times to generate the 100 Swifties. The first attempt was over 50% male, which did not pass the sniff test. After some iteration, I landed on a distribution that felt closer to reality (still not perfect, but more “Eras Tour stadium” and less “Buffalo Wild Wings on a Thursday”).
The App
Built in Replit, using their new improved model + extended features. Honestly one of the cheapest apps I’ve ever made: I thought I was done at ~$9, but feature creep (aka the survey functionality) pushed it to around $20. I wasn’t satisfied with just “chatting with three Swifties at a time” — it felt like slow fandom speed-dating — so I built the survey mode so you could query the whole dataset at once.
The Model(s)
This was also my first time using the GPT-5 API in a project. I started with GPT-5, then swapped to 5-nano for speed. It’s faster, but not dramatically so. If I ever tried to replicate this for a noble cause, I’d definitely try to consolidate API calls — right now, each “Swiftie” response is its own call, which is hilariously inefficient. Still, GPT-5 is so cheap that I haven’t even registered a single cent of token spend yet.
Color Palette
Midway through, I used ChatGPT to help me design an Eras Tour–inspired color palette for the app UI. Because if you’re going to simulate fandom, you might as well make it look good.
Security (or lack thereof)
I didn’t bother putting authentication in front of this. So technically anyone can see any poll results. This is for fun, not science, and I’ll probably unpublish it eventually. Consider it a pop-up museum of unnecessary code.
It is asynchronous but does take a bit so fire off your survey and come back after a drink or snack