Customer support · Production

Google Research Wayfinding AI agent uses proactive clarifying questions to deliver more helpful personalized health information

The problem

Navigating online health information is confusing and impersonal, and most AI tools act as passive question-answerers that provide generic comprehensive answers without seeking the context needed to give personalized guidance. People also struggle to articulate their health concerns without a clinical background.

Workflow diagram · grounded in source
1
User submits health question
trigger
“participants were instructed to have a conversation spending at least 3 minutes on their question”
2
AI asks clarifying questions
ai_action
“by proactively asking clarifying questions, an AI agent can better discover a user's needs, guide them in articulating their concerns, and provide more helpful, tailored information”
3
Two-column interface delivers response
output
“we designed an interface with a two-column layout. The conversation and clarifying questions appear in the left column, while best-effort answers and more detailed explanations appear in the right”
Reported outcome

Users preferred Wayfinding AI over a baseline Gemini 2.5 Flash model across helpfulness, relevance, goal understanding, and tailoring.
Conversations were longer, with 4.96 turns on average versus 3.29 for the baseline on symptom-related topics.

Reported metrics
average conversation turns - Wayfinding AI (symptom topics)4.96
average conversation turns - baseline AI (symptom topics)3.29
Total study participants across four studies163
Randomized user study participants130
Show all 6 reported metrics
average conversation turns - Wayfinding AI (symptom topics)4.96
average conversation turns - baseline AI (symptom topics)3.29
total study participants across four studies163
randomized user study participants130
user preference for deferred-answer approachmajority of participants preferred
user helpfulness, relevance, and tailoring rating vs baselinesignificantly more helpful, relevant, and tailored
Reported stack
GeminiGemini 2.5 Flash
Source
https://research.google/blog/towards-better-health-conversations-research-insights-on-a-wayfinding-ai-agent-based-on-gemini/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Users preferred Wayfinding AI over a baseline Gemini 2.5 Flash model across helpfulness, relevance, goal understanding, and tailoring.

What tools did this team use?

Gemini, Gemini 2.5 Flash.

What results were reported?

average conversation turns - Wayfinding AI (symptom topics): 4.96; average conversation turns - baseline AI (symptom topics): 3.29; Total study participants across four studies: 163; Randomized user study participants: 130 (source-reported, not independently verified).

How is this customer support AI workflow structured?

User submits health question → AI asks clarifying questions → Two-column interface delivers response.

Related customer support cases