Netflix tests AI Discover tab to reduce aimless TV scrolling

Netflix is adding an AI-powered recommendation tool to its TV app, starting with theme-based suggestions and explicit feedback controls. The test signals a shift toward user-directed AI discovery.

What happened

Netflix is adding an AI-powered recommendation tool to its TV app as it tries to get viewers to stop scrolling so much and watch more shows and movies. The company framed AI Discover as a testbed for new features that it might make part of the core Netflix experience. If it's available on your account, you'll see a Discover tab with a red gem icon on the top navigation bar.

The first thing Netflix is trying out in AI Discover is theme-based recommendations. The tab displays an array of themes along with a relevant clip from a show or movie. Themes are based on who you might be watching with, such as a family movie night option; what you're in the mood for, like a horror flick or light-hearted comedy; or titles that won't take up your entire evening. The themes can also include new and/or popular shows and movies, as well as recommendations based on what you've been watching recently.

After picking an option, users can refine the recommendations with more specific themes and refresh the selection. They can tell Netflix not to recommend a title again, which should help refine recommendations and block unwanted titles, and they can add titles to their list from AI Discover.

"AI helps create the themes you see, and we'll explain your results along the way," Netflix said. "AI Discover is powered by a new intelligence layer we've been building that responds to what members tell us in real time and keeps learning."

Key facts

• AI Discover is initially available on supported TVs in English in the US, with gradual rollout in the coming weeks.
• The tab will not appear on kids profiles, according to Variety.
• Theme signals include co-viewing context, mood, viewing time, new or popular titles, and recent watch history.
• Users can block a title from being recommended again and can add titles to their list directly from AI Discover.
• Netflix previously tried a shuffle play option a few years back, tested an OpenAI-powered chatbot when it rolled out a redesign last year, and once had the Max tool over a decade ago.

Our analysis

This is a meaningful shift from passive recommendation rows to a more interactive discovery loop. Instead of only inferring preferences from watch history, Netflix is asking viewers to pick a context and then refine it. The "new intelligence layer" that responds to what members tell it in real time suggests that direct feedback will become a stronger ranking signal, not just a secondary data point.

For content teams, the relevant clip attached to each theme is likely to become a primary discovery asset. A single preview has to communicate tone, mood, and watchability before a title is selected. The ability to say "don't recommend this again" creates a hard negative signal that earlier thumbs-down or skip controls may not have provided, meaning content that irritates users could be suppressed more quickly in future recommendation surfaces.

Because Netflix calls AI Discover a testbed, features that work here may influence the core Netflix experience and could shape how other platforms design AI-driven content discovery. The explicit theme categories—mood, viewing time, co-viewing—suggest that content metadata and preview strategy will matter more for algorithmic distribution.

What it means for operators

• Treat short clips and preview thumbnails as the primary discovery asset, not an afterthought, because AI Discover pairs each theme with a relevant clip from a show or movie.
• Build content around clear mood, time commitment, and co-viewing signals so titles can match theme-based recommendation categories.
• Watch how explicit negative feedback controls are exposed; if a platform adds a "don't recommend again" option, reduce content patterns that trigger immediate dismissal.
• Monitor AI Discover and similar testbeds as signals for where recommender systems are heading, then adapt metadata, packaging, and preview assets before those features expand.