Imagine you have a friend who has seen basically every movie or TV show ever made. They’ve read every piece of analysis. They know the plot, the themes, the sub-themes, and the sub-sub-themes.

They’d probably be pretty good at six degrees of Kevin Bacon. Sure. But they’d also probably be pretty good at seven degrees of your taste. Given thing you liked A, and thing you liked B, what else might you like?

The hard part for them isn’t knowing the path from A to B, the hard part is knowing A and knowing B.

Well, I have good news. That friend’s name is Claude. And here’s how I taught it my own personal point A and point B across various kinds of media.

Step 1

Dear Claude, hi how are you? I’m fine. I would like you to interview me about my taste in movies. I think a good way to explore this is by talking about what movies I’ve liked and why I liked them, as well as movies I didn’t like, and why I didn’t like them. My goal is to have you write a markdown file that captures my taste preferences in movies, not as a list of movies I’ve liked, but as the themes and “clumps” that run across them, and which could be used to predict movies I’d like in the future.

Step 2

I don’t type very much anymore. In fact nearly the only exception is blogging, where I find the act of typing it out with my own meaty paws essential to the thinking process. But for everything else I dictate. Voice to text has gotten more than good enough, and I’ve always loved to yap.

So this is where I’d start a nearly 15-minute-long ramble going through every movie I could think of. When I did this for real, I had two things that helped.

First: I started recording every movie I watched around the start of the year to feed a year-end “Wizardly Wrapped” (coming Soon™).

Second: I digitized my movie library, so it’s easy to scroll through the list.

I did both, rambling without any structure into my VTT app of choice (Every’s Monologue, which I like because it can replace the iOS keyboard, but I’ve used others and they’re all fine-or-better), and dumping that unstructured thought directly into Claude’s prompt-box.

Claude is extremely good at divining signal from noise, especially stream-of-consciousness.

Step 3

Claude mined it for themes, and produced a long document that included this:

  • Big-canvas sci-fi and fantasy. Space, deep time, strange new worlds. Earnest and cerebral works as well as spectacle: Arrival, Aniara, Project Hail Mary, Armageddon.
  • Glossy period ensemble and cozy mystery. Costumes, wit, a house full of suspects or sisters.
  • Camp and gothic with a wink. Chucky, Hellraiser, Austin Powers, the Addams register. Serious gothic horror on the Eggers end of the same lane. He likes horror and rates it modestly: Hellraiser II is a 5 and still “solidly fine.”
  • Earnest, heartfelt, a little whimsical. The Joy Luck Club, Everything Everywhere All at Once, Stranger Than Fiction, K-Pop Demon Hunters, Kubo, Pixar. Asian and Asian-American stories are a real axis here, not a coincidence: three of his top five films of 2026.
  • Artful animation. Laika, Pixar, Ghibli, Sony Animation. The line is the studio and the ambition, not the medium. The Bad Guys yes; Angry Birds 3 and Bluey no.
  • Musicals. Movie musicals and stage recordings of musicals he would love (Hamilton, Come From Away). Music can carry a film for him: the K-Pop Demon Hunters soundtrack ran for two weeks.

Which. Yeah. Pretty on the nose.

Step 4

So how do you operationalize your new taste profile? For me, it depends. I use them to filter recommendations for music, TV, or books. “Hey Claude, would I like this?” Now it has a basis for knowing whether or not I would.

For movies, I added something to my daily newspaper (more on that in a future Adventures in Claude) that sweeps ahead for new movies to see if there’s anything I might want to see. The answer most of the time is no. I’m not a big movie person. But that also means I’m totally keyed out of trailers and have no other system that might reliably let me know that Dune 3 tickets are on sale.

Step 5

Do it again. For anything where it might be useful for the AI of your choice to understand your tastes and preferences. I have found this is often the first hurdle for new people trying to get AI to “do something useful.” As capable as they are, they don’t know you. They don’t know what you want, or how you work, or what you’d consider helpful or annoying.

So teach them. It’s as easy as talking about the things you love, something we should all do more often.