Designing AI-powered search for millions
When Zillow, Homes.com, and Jitty launched AI search, Realtor.com had eight weeks to respond. I led a cross-platform sprint that rebuilt fragmented filters, enabled natural-language search, and laid the foundation for AI-powered discovery. The work shipped ahead of our October 2025 AI campaign. Within two weeks, traffic from search results to listings grew by 12%.
Company
Realtor.com
Year
2025
Role
Lead Product Designer
Timeline
8 weeks
01 CHALLENGE
Our filters hadn't evolved in years—cluttered, inconsistent, barely used. Analysis of 93M monthly users revealed only 13% applied any filters beyond location, with engagement concentrated in just 5 core options. Most people avoided filters entirely and scrolled through pages of irrelevant homes.
User research revealed why. The problem wasn't the data—it was fear. Fear of filtering themselves out of a great home. This fear drove every design decision that followed.
The core issues:
Only 13% of users applied filters—most avoided them entirely
Inconsistent naming and controls across iOS, Android, and web eroded trust
40+ rarely-used filters created decision paralysis
No support for soft preferences like "prefer a pool"
Long-tail filters used by under 0.5% of users wasted prime screen space
To launch AI search, we first needed to fix the foundation.
02 RESEARCH
I started by analyzing filter usage data across 93M monthly users. Only 13% of users applied filters, with engagement concentrated in just 5 core options. The remaining 35+ filters created noise. User research revealed why: filtering felt like a gamble. Filter too much and you miss homes. Filter too little and you waste time. Most preferred browsing broadly over trusting the system.
This insight drove everything that followed: users wanted natural language search, not more filter options. But to enable AI to interpret conversational queries, I first needed to fix the fragmented foundation.
Cross-Platform Audit
I audited 40+ filters across web, iOS, and Android, documenting usage rates and naming inconsistencies. The findings were clear: core filters like property type saw 63% usage, while advanced options like commute distance barely reached 0.5%. We were giving equal visual weight to filters that had wildly different value to users.
I found major inconsistencies: "Garage" lived in different categories across platforms. "Property Details" and "Property Features" meant the same thing. These problems would break AI's ability to interpret natural language queries.
Competitive Landscape
I analyzed 13+ competitors—Zillow, Redfin, Jitty, Homes.com, and cross-industry leaders like Airbnb—across 8 criteria. The best experiences prioritized clarity and progressive disclosure over comprehensive filter lists. Critically, platforms with AI search relied on unified taxonomies to interpret natural language queries. The analysis revealed I couldn't launch AI search on top of fragmented filters. The foundation had to come first.
The analysis revealed I couldn't launch AI search on top of fragmented filters. The foundation had to come first.
Building Alignment
I brought Product, Engineering, Research, and Design Systems together for a one-day workshop. We synthesized research findings, audited filter behavior across platforms, and defined 7 core principles:
7 Core Principles
Create focus and de-clutter
Prioritize what's most useful
Improve hierarchy of filter groups
Make filters easier to scan and understand
Use controls that match intent
Achieve platform parity
Keep patterns consistent
These principles became our shared language for making decisions under pressure. With everyone aligned, I moved into an 8-week sprint to rebuild the system.
03 SOLUTION
I broke the work into three parallel tracks that would come together to enable AI: unified taxonomy, redesigned interface, and scalable systems. Each track had clear success criteria and interdependencies, allowing multiple teams to move fast without blocking each other.
I ran two rounds of stakeholder reviews—first with cross-functional teams, then with executive leadership. These kept everyone aligned on timing, validated design decisions, and helped sequence delivery across platforms.
Unified Taxonomy
I rebuilt the filter taxonomy from the ground up—consolidating sections, cutting rarely-used options, and standardizing labels across platforms. Duplicates like "Waterfront" and "Lake View" became a single filter. The unified structure let AI map natural language queries like "three-bedroom home with a pool" directly to our filters.
Redesigned Interface
I replaced checkboxes with chip-based inputs for faster scanning and clearer states. Added multi-select for listing status. Simplified bed and bath filters to match how people actually search. Every control was designed with accessibility in mind, ensuring the experience worked for all users. The cleaner interface created space for AI features—users could quickly refine AI-generated results without visual overload.
Scalable Systems
I partnered with our design system team to evolve the component library. We expanded the color palette for better hierarchy and accessibility, refined typography for improved scannability, and ensured every pattern met accessibility standards. I aligned design tokens, patterns, and components across web, iOS, and Android—creating platform parity so users got the same experience whether they started on mobile or desktop.
These components became the foundation for rapid AI feature development. Engineering could focus on AI logic instead of rebuilding filter interactions from scratch.
04 impact
In eight weeks, I delivered a complete redesign of Realtor.com's search across iOS, Android, and web. The foundation powered the company's first AI search experience, launched ahead of the October 2025 AI campaign.
05 REFLECTION
This project taught me how to turn urgent requests into lasting systems. Under pressure to ship fast, I resisted the urge to patch. I fixed the foundation instead, and that made future work easier.
The filter system I built will support AI features for years to come. The workshop created alignment across four teams that still holds today. The guiding principles became our shared language for making decisions quickly.
Biggest lesson: Urgency can drive clarity. When you only have eight weeks, you focus on what actually matters. Speed requires strategy, not shortcuts.