The Problem with Movie Recommendations
Netflix knows what you’ll watch next. It doesn’t know why, and it can’t explain the recommendation in a way that helps you discover what you actually want. Algorithms optimize for click — not for the kind of film that stays with you for weeks.
The Approach
Over several months I fed ChatGPT a structured log of every film I’d watched and rated (200+). The prompt evolved from simple “recommend me something” to a much richer dialogue:
- Rating (1–10) + Why (2–3 sentences on what worked and didn’t)
- Mood at time of watching
- Rewatch factor — would I watch it again, and in what context?
- Taste axis — where does this sit on the slow/fast, cerebral/visceral, dark/hopeful axes?
From this, ChatGPT extracted a personal taste fingerprint: strong preference for morally ambiguous protagonists, slow-burn pacing with earned payoffs, cinematography-first filmmaking, and anything in the noir/neo-noir family.
Book/Genre Discovery
The same system revealed two blind spots: hard sci-fi (I’d been avoiding it; after profiling my taste it identified specific authors who’d bridge the gap) and Nordic noir (which led to a reading list that replaced my Netflix habit entirely).
The AI also identified that I respond strongly to audio drama as a format — long-form narrative podcasts and BBC Radio plays. This wasn’t something I would have articulated myself; it emerged from pattern analysis across my ratings.
What I Track Now
- Film ratings + taste notes (ongoing log)
- Book completions + whether I’d recommend and to whom
- Audio drama favorites (BBC Sherlock Holmes, Audible Originals, Wolf 359)
- Genre evolution over time — where my taste is heading
The system has become a genuine personal media intelligence layer. I ask it “what should I watch tonight given I want something like X but haven’t seen Y” and get a ranked list with a paragraph of reasoning for each choice.