How can we improve the current Copilot Local Answer?
I partnered with a product manager at Microsoft to explore concepts for an improved answer in Microsoft Copilot.
Project type
Conceptual explorations
Skills
Design explorations, prototyping, research application
Team
1 product manager, 1 designer
Role
Product Designer
Context
When you ask Copilot for "restaurants near me", it returns a simple answer.
The current "Copilot Local Answer" displays limited information and doesn't allow users to dive deeper into their options. Based on past research research, these are core parts of a local search journey, so we needed to design a way for users to do this within Copilot.
Opportunity #1
Make it easier for users to compare their options
Observations
Local queries like "restaurants near me" signal comparison mode.
Users want quick ways to understand differences between options.
Copilot users expect AI to interpret, not just list information.
Insight
If AI can distill signals from across the web into clear, trustworthy highlights, users will gain a deeper understanding of each option with less effort.
This layout prioritized a default comparison view, surfacing basic details alongside AI‑generated highlights to help users quickly evaluate their options. For users who prefer spatial exploration, it also offered a map view as an alternative entry point.
Opportunity #2
Enable Users to Dive Deeper
Observations
Prior research on Local decision journeys showed that once users begin narrowing down their choices, they need access to richer, entity-level details to feel confident.
The most commonly referenced information included: ratings and reviews, photos, price ranges, hours of operation, and menus (for restaurants)
In the current design, users could see only high-level summaries. There was no clear path to dive deeper into individual entities, which created friction at a critical decision point.
Insight
Users need a structured, lightweight way to access deeper details without losing their place in the broader results.
Exploration 1
Chat-first disclosure
This approach assumed that the most natural pattern for Copilot is to keep all interactions within the chat paradigm. Instead of relying on clickable UI to reveal more information, users would simply ask Copilot for deeper details, and the chatbot would surface them directly in the thread. We leaned toward this option because it stayed truest to the core expectations and mental model of a chat-based experience.
Exploration 2
In‑context details panel
This approach displayed detailed entity information in a right‑hand details panel, allowing users to remain in the familiar chat surface while viewing additional information in a fixed, dismissible space. The goal was to support deeper exploration without disrupting the flow of the conversation.
Exploration 3
L2 details page
This approach surfaced richer entity information on a dedicated L2 page, where users could continue chatting with Copilot to ask follow‑up questions. However, we questioned whether introducing an L2 pattern aligned with Copilot's core interaction model. Because users are accustomed to staying within a single conversational surface, adding a secondary page risked feeling heavy‑handed and potentially out of character for a chat‑first experience.
Outcomes
Although the initiative was paused before moving beyond MVP, the V2 explorations created meaningful impact across the Copilot product ecosystem.
Influenced adjacent teams
Several of the patterns I developed independently aligned with explorations happening in other product segments. These patterns were later adopted, contributing to a more unified and scalable design direction.
Clarified the long-term vision
The work established a clear north star for how the experience could evolve, helping the team articulate what a more extensible and user-centered version of the feature could look like.
De-risked future decisions
By exploring multiple paths early, the work surfaced usability, technical, and architectural considerations that will inform future iterations once the project resumes.