VladMarchenko
Résumé Let’s talk
Case · Kettering Health

AI search that works out what a patient means

A new search and results layer for Kettering Health, a hospital network in Ohio. It had to fit inside the existing site and design system.

ketteringhealth.org
The new result widgets

Patients typed symptoms and worries. Search answered with lists. I redesigned it to answer first and point to a next step.

Problem. Search only matched keywords. It worked if patients already knew what to type.

Outcome
38%

fewer searches that ended with no click

2.4x

more appointments started from search

12s

median time to the first useful result

5

hospital sites on the new layer

Projected
My part
Owned the AI search logic end to end: query intents, keyword rules, widget ranking and scoring, and safe answers for personal health questions. Presented every decision to the client and their marketing agency.
Team
Me, a design lead and a project manager, with Kettering’s marketing agency
Year
2025
The old search results page for “knee pain”: a keyword list of related content
The new results for “knee pain”: Quick Assist answers first, with specialists, locations and related content
Before / after

Same search, new answer

The same search for knee pain, on the old page and the new one.

01 · Visual overview

A fresh layer on a trusted system

I kept the existing system and added one pink accent for AI, plus soft gradients.

The new search had to live inside an established design system, so consistency came first.

Within those limits I added what makes the AI feel new without breaking familiarity: a refined colour code with a dedicated pink for AI features, and subtle gradients. The result stays stable and still looks forward.

Results page on a laptop
Guided search on a phone
Quick Assist card
03 · Action-oriented results

From queries to clear actions

Lists of links became widgets that each stand for a next step.

Once intent is clear, results have to feel useful, not overwhelming. Before, patients saw long lists of links; even when the right option was there, it was lost in the noise.

I replaced static lists with modular, action-oriented widgets. Each one is a clear next step — book an appointment, find a specialist, check a location, read up on a condition — so patients act instead of interpreting raw data.

Follow-up prompts adapt to the context, like narrowing a location search to “closest to me” or widening a treatment query. Patients always see a way forward, whether they're ready to act or still deciding.

Result widgets: doctors, locations, events
Results on a phone
All related content list
Related suggestions
04 · Clear and supportive

Designing for confidence and clarity

AI answers in plain language, and clarifying questions instead of “no results”.

Health searches can be stressful, especially when people aren't sure how to describe what's wrong. Long, equal-weight lists only added to the uncertainty.

I highlighted the details that matter most, wrote the AI answers in plain language and replaced “no results” with clarifying questions. On mobile, results appear instantly with fewer steps.

The aim: less anxiety, and a calmer path from concern to action.

Quick Assist answers for three different searches
A provider found from the search on a phone
05 · Healthcare-ready

Seamless integration with healthcare systems

Only the search layer changed; booking stays in the existing HIPAA-compliant system.

I focused on the search and results layer and left booking and patient records untouched. Patients preview slots, check locations and explore services in the new design, then finish scheduling in the existing HIPAA-compliant system.

That kept the rollout fast, the risk low and compliance intact.

A sensitive question answered with care on a laptop
Search Assist explanation on a phone
Verified sources under an answer
Looking back
  1. 01
    Simpler beat complete

    We first tried to cover every intent with ranked widgets and rules for which ones could show together. The final design shows two best guesses, then the full results list. Covering everything isn’t the same as helping.

  2. 02
    Push for the start of the journey

    Search starts on the home page, but our scope was the overlay and the results. I proposed taking the home page too, and the client said no. Next time I’d push harder, with data. I prefer owning a flow from start to end.

  3. 03
    Trust AI more, then check it

    It was a health project, so I kept AI on a short leash and did most of the analysis by hand. Now I’d give it much more of the work and put that caution into checking the results.

Want the full story?

The decisions behind each screen, how the answers were checked, and what I'd push further next. Happy to walk you through it on a short call.

01Visual overview
the story