amanda severianoProduct Designer
All projects
Client
Large real estate developer (via Igma)
Year
2026
Role
Product Designer
Goal
AI on the website + better-qualified leads

AI search: from filtering by city to searching by lifestyle

  • Discovery
  • Market analysis
  • Experience rules
  • Functional prototype
  • Pilot
  • 1 sprintfrom discovery to prototype
  • 5platforms analyzed
  • 83.9%of leads come from mobile
  • 6experience rules
AI search on the developer's website: the person types 'apartment with 2 bedrooms and a balcony in BH to buy' and the AI shows what it understood as editable tags above the results.

A large real estate developer's website search only worked by state and city, and it failed silently: no error data, just people leaving the site and leads the brokers discarded.

At the same time, the developer wanted to finally deliver a two-year-old request: an AI chat on the website, also as a sign of investment in technology. The chat required more than the team had, so we proposed a middle path.

The request

AI chat

It offered

  • Conversational, step-by-step guidance inside the website.

Why it didn't fit

  • It required more effort, time and risk than the team had available.
Delivered

AI search

What we gained

  • AI on the website, within the team's capacity.
  • Data, for the first time, on what people search for.
  • Less risk of a slow experience, like Redfin's chat.

What we gave up

  • The conversational guidance.

A natural language search: it delivers real AI, without a conversation. The person describes what they're looking for, the AI shows what it understood as editable tags and never leaves them without an answer.

BeforeSearch results by state and city. A screen with misaligned content.
Before: search results by state and city, with misaligned content.
AfterThe search understands "2 bedrooms and a balcony in BH to buy".
After: the natural language search shows tags for what it understood and the homes found.
AI search results: 'We found 123 homes for you', with the understood filters (Rent, House, Filters) next to the field.
No results state: 'We're not in this location yet', with the options Notify me and See nearby.

Since the problem was silent and there was no data on the current search, I analyzed 5 market platforms to validate the premises:

DecisionThe AI shows its interpretation and each criterion becomes a removable tagWhy?
PremiseShowing what the AI understood builds trust
What the market showedZillow, the only one with established natural language search, turns the text into visible filters.
DecisionDirect input and response, no conversationWhy?
PremiseChat tires out people who just want to see homes
What the market showedRedfin's chat makes searching slow and tiring.
DecisionAI search gets visual prominence on the home pageWhy?
PremiseHidden AI doesn't get used
What the market showedOn Redfin, smart search sits in the background and loses usage.
DecisionThe old search stays available, next to the new oneWhy?
PremisePeople who already use the site can't lose out
WhyReturning visitors are used to searching by state and city. Removing that path would break a habit that worked for them.
DecisionBottom sheet as an MVP requirement, not a future improvementWhy?
PremiseOn mobile, the bottom sheet is the standard
What the market showed5 out of 5 platforms use a bottom sheet, and 83.9% of leads come from mobile.
OpportunityVoice search, still unseen in the marketWhy?
What we sawNone of the 5 platforms analyzed offers voice search.
Why it mattersOn mobile, where 83.9% of leads come from, speaking is easier than typing.
OpportunityInstallment, condo fees and property tax within the AI's reachWhy?
What we sawBrazilian platforms already show condo fees and property tax on the card. International ones don't.
Why it mattersWith these costs linked to the properties in the backend, the AI can answer "installments up to R$ 1,500", and the lead reaches the broker better informed.

A business constraint

The property card and page were out of scope, by business decision. Showing more information there had already caused sales problems, because a card doesn't sell a single home: it sells different floor plans in the same development. So lifestyle and prices wouldn't be shown to people searching.

The constraint

No changes to the property

What was left out

  • Lifestyle and prices on the property card and page.

Why

  • A card groups different floor plans in the same development. Too much detail had already caused sales problems.
My counterproposal, accepted

Data in the backend

What we gained

  • Lifestyle and prices registered and searchable by the AI.
  • The search filters by what the person actually wants, even without showing it on screen.
  • The lead reaches the broker better qualified, even if they still have questions.

What we gave up

  • Showing this information to people searching.

I also put it on record with the team: I see value in rethinking how properties are registered and managed, involving other teams if needed.

AI search on mobile, screen 1.
AI search on mobile, screen 2.
AI search on mobile, screen 3.

Where we are

  1. DoneDiscovery
  2. DoneRules and prototype
  3. ● In progressPilot in a few cities
  4. ○ NextValidate and adjust
  5. ○ NextScale, if positive

Learnings

  • Cutting the scope from a chat to a search made it possible to deliver real AI within what the team could do.
  • A silent problem needs instruments before solutions: I would measure the current search from the start, to compare before and after.
  • When the business locks the interface, you can still win on the data: what doesn't show on screen can work in the search.

Shall we talk?

I like turning ideas into simple experiences. If you want to talk about design, research or opportunities, I'd love to hear from you.

Message me on WhatsApp