Netflix Is Reportedly Testing an AI-Powered Voice Search Feature

Netflix is reportedly testing a native artificial intelligence (AI)-powered voice search feature. As per the report, the California-based streaming giant is building a semantic search system that can recommend and surface content based on user intent and contextual cues. The feature is currently available in beta and is said to be accessible to only a select group of users. It is unclear when the company might release it for all users, or if it plans to reserve it for specific tiers.

Netflix Reportedly Testing AI Voice Search

According to The Verge, the streaming giant has rolled out the AI voice search in beta to a limited number of users in the US. It is said that the capability is currently only available on a specific device/platform combination. The publication said it worked on a “Chromecast with Google TV streaming dongle and a TCL Google TV.” However, the AI experience was reportedly not available on Roku or Fire TV devices.

Once the Netflix feature is activated, users are reportedly shown a few search suggestions, such as “watch it in the background” and “I need a good cry.” Selecting these options is said to open up a set of recommended content. The report mentioned that there is also an “Ask” button with a waveform icon that lets users request movies and web series based on their mood or any random criteria, such as “movies after a long, tiring day.”

The underlying large language model (LLM) is said to be capable enough to understand the intent of the user and present them with accurate suggestions, even for obscure prompts. For instance, the report mentioned that when the voice search feature was stress-tested with “fun kids TV shows about death,” the AI surfaced A Series of Unfortunate Events and Raising Dion, both of which fit the description perfectly.

There are some limitations to the beta version of the Netflix feature. The report mentions that it cannot connect to the platform's personalisation engine and cannot suggest content based on a user's watch history. It reportedly also does not talk back to the user and can only respond via text. Finally, the AI search is also said to be prone to making some mistakes when surfacing content, but those instances were said to be rare.