Netflix-Style AI Profiles Work Best Only for Binge-Curious Viewers

Streaming services decide what to recommend using a 'user profile' — basically a snapshot of what you've watched. Researchers from a production streaming platform compared two versions. The first is the classic approach: a statistical summary built from your viewing history. The second uses a large language model — the same kind of AI behind chatbots — to read your history and write a short natural-language description of your tastes, like 'likes moody Scandinavian thrillers and cooking shows.'

The surprising result: the fancy AI version wasn't better for most people. For roughly four out of five users — those with steady, habitual viewing patterns — the old-fashioned profile gave better recommendations. The AI summary only pulled ahead for 'exploratory' users: people who jump between genres and whose next watch doesn't look much like their past.

There was a second catch. The AI-generated profiles had a 'popularity attractor' effect. In plain terms, the AI kept gravitating toward hit shows, the way a friend might keep recommending whatever's trending. That made each individual recommendation list feel a bit more varied, but it shrank how much of the full catalog got surfaced and made suggestions less novel. You'd see more of the same blockbusters and fewer hidden gems.

The takeaway for platforms: don't use one method for everybody. The researchers suggest a smarter system that first figures out what kind of viewer you are — routine or restless — and then picks the matching technique. For you, that could eventually mean recommendations that actually fit your mood instead of guessing. But it also means AI profiling of your habits is spreading, and it can quietly narrow what you're shown.