Adding AI to Search
Senior Product designer (Team Lead)
Charter - Discovery | Pod - Search
with Somanshu: Associate Product Designer
Client:
Zee Entertainment
Year:
2022
The Challenge
Search was a critical but stagnant flow—unchanged for years due to its impact on a key metric: Search CTR.
High oversight — every change required business approval
Underutilised insights — strong data existed but wasn’t translated into product
Low iteration — risk aversion limited experimentation
Underperformance — CTR at 73% vs ~80% industry benchmark
Impact
Redesigned and launched AI-powered search
Improved Search CTR from 79% → 81%
Operationalised data and research into product decisions
Influenced multiple downstream features (e.g. language rails, content formats)
Brought previously unprioritised user problems onto the roadmap
Approach
I used a hackathon as a wedge to bypass approval constraints and prototype a new direction.
The initial proposal was a chat-based interface or natural language search search based on past failures and user expectation (Jacobs law), I challenged this direction. Alternatively I designed a tappable prompt component, subtly educating the user about long form text being available instead of just keywords.
We built the search experience in a record time of 2 days.
Based on the hackathon success, leadership wanted to launch to production,
Key Constraints & Decisions
AI search was gated behind an opt-in entry (banner) to avoid business exposure
I pushed for default entry or toggle to improve discovery (top-of-funnel), but this was deprioritised due to dev constraints
This trade-off directly impacted feature adoption—validated later in research.
Feature was launched to 5% of iOS users as an experiment
Phase 2
Strong early signals led leadership to expand the feature scope:
Merge AI and standard search into a single experience
Introduce new content rails to drive consumption
Enable advertising and branding opportunities
Reduce result latency
Create a distinct visual identity for search
Make the experience feel more conversational
The PM scoped delivery screen-by-screen to optimise development bandwidth. I chose to design the end-to-end flow to ensure a coherent experience.
I defined the landing page structure while Somanshu handled its detailing.
In parallel, I partnered with the data analyst to surface long-standing user issues and combined them with research insights.
This allowed me to go beyond the initial scope and solve for validated user problems across the results and typing states. The solution was well received and successfully added to the roadmap.
svdevdev
User Testing & Validation
Conducted qualitative research with 18 users in partnership with the research team.
Key insights:
~50% (9/18) exited the app to search externally
Large, unstructured results increased confusion
Users defaulted to exact keyword search
AI search was misunderstood due to mental model gaps
Tappable prompts were appreciated but misinterpreted
Outcome:
Grounded data insights in real user behavior, enabling us to design solutions for actual user needs rather than assumptions.
User testing results
Data & insights
Partnered with data to analyse cross-platform search (Android, CTV, iOS, Web), uncovering high watch time but lower CTR for SVOD users.
Insight → execution — Designed high-fidelity flows for contextual search and non-serialized content rails to address poor relevance in results.
Roadmap influence — Drove 12 data-backed recommendations into the roadmap, including multi-language standardisation and out-of-catalogue handling to reduce drop-offs.
Research by data analysts
Recent search retained due to high conversion
New rail experiments based on user feedback
svdevdev
Loading state animation
Results screen hero component
Social proofing
Top searched for episodic content
Spelling errors fixed
Content now found error experience fixed