Rabbit AI | Search & Discovery
Tubi
Overview
At Tubi, I worked on Rabbit AI, an AI-powered experience designed to help viewers discover what to watch.
Traditional streaming search worked well when users knew exactly what they wanted, but struggled when they searched using moods, preferences, or vague descriptions.
The biggest challenge was to make discovery more conversational and intuitive, helping users find relevant content without knowing specific titles or genres.
My role
I partnered with product, engineering, and AI teams to explore how large language models could improve content discovery.
We initially explored a standalone conversational assistant, but I saw an opportunity to integrate AI more directly into the existing search experience.
My focus was on three principles: understand natural-language intent, make recommendations relevant and explainable, and help viewers move from exploration to watching with minimal friction.
Key Impact
Rabbit AI drove 18% more search refinements and 12% higher click-through rates.
Subsequent iterations reduced measured churn from 56% to 39%, improving content discovery and engagement.
Search Query Conversion
New Results Discovery
The Problem
Traditional streaming search assumes viewers know what they want to watch. But many viewers arrive with only a mood, an interest, or a vague idea, making keyword-based discovery frustrating and limiting their ability to explore Tubi's extensive content library.
With over 100,000 titles available, finding relevant entertainment could require repeated searches and browsing. We saw an opportunity to make content discovery more intuitive by allowing viewers to describe what they wanted in their own words.
The Opportunity
Large language models introduced a new way to interpret viewer intent beyond exact titles, actors, and genres. We explored how conversational AI could translate open-ended requests into relevant recommendations, making content discovery more personalized and accessible.
The product opportunity wasn't simply to introduce a chatbot. It was to reduce the distance between a viewer's intent and something worth watching.
Research & Discovery
I collaborated with cross-functional partners to examine limitations in existing search behaviors and explore how viewers might interact with conversational recommendations. We considered ambiguous searches, mood-based requests, contextual preferences, and the friction of moving between conversation and content selection.
These explorations informed three design principles: understand intent beyond keywords, make recommendations easy to evaluate, and minimize friction between discovery and playback.
Key Design Decisions
Conversation vs. Search — Balanced the novelty of a standalone chatbot with the familiarity and efficiency of existing search interactions.
Exploration vs. Selection — Considered how to encourage conversational refinement without making the journey to playback unnecessarily long.
AI Flexibility vs. Predictability — Focused on helping users understand and act on recommendations, rather than making AI capabilities the center of the experience.
Experimentation & Iteration
We used early product experimentation to learn how viewers engaged with conversational discovery and where the experience introduced friction. Findings from the initial release informed subsequent improvements to recommendation relevance, search refinement, and interaction patterns.
Rather than treating Rabbit AI as a single launch, we approached it as an evolving capability that could improve how viewers discover content throughout Tubi.
Outcomes & Business Impact
Rabbit AI demonstrated the potential for conversational AI to improve content exploration, generating an 18% increase in search refinement and a 12% lift in click-through rates. Subsequent iterations saw measured churn decrease from 56% to 39%, suggesting meaningful improvements in the discovery experience.
Beyond engagement metrics, the project provided a foundation for exploring how generative AI could complement Tubi's broader personalization and discovery ecosystem.
Reflection & Learnings
Rabbit AI reinforced that successful AI products begin with user needs, not the novelty of the technology. The most valuable opportunity wasn't creating a chatbot, but helping viewers express their intent naturally and discover content with less effort.
The experience strengthened my approach to product strategy: validate emerging technology through focused experiments, challenge assumptions about interaction models, and use behavioral evidence to guide the next iteration.