This project is password protected

Emily Thomas
Google

Workspace Persistent Conversations
Workspace Gemini Standardization

As Gemini rolled out across Workspace, a critical issue emerged: the same AI behaved differently across products and surfaces—creating confusion and breaking user trust. I led the research and design exploration to define how Gemini conversations should persist across Workspace products, establishing a framework that would enable seamless, cross-product AI workflows.

Cross-product Gemini conversation flow
My Role

Lead UX

My Team

1 Researcher

My Contribution
  • Identified the opportunity to persist Gemini conversations across Workspace, reframing it as a system-level UX problem
  • Defined and evaluated three conversation models, building a high-fidelity prototype to enable A/B/C testing
  • Established the long-term product strategy and secured VP alignment

Context

Business Goal

In 2025, Google Workspace set out to lead in AI-powered productivity by introducing an assistive AI side panel across products. The goal was to enable users to seamlessly find, analyze, and act on content using Gemini—no matter where work happens.

Insight

As Gemini rolled out across Workspace, a critical issue emerged:

The same AI behaved differently across products and surfaces—creating confusion and breaking user trust.

Users expected a consistent, continuous AI experience, but instead encountered fragmented interactions depending on where they were (Drive, Gmail, Docs, etc.).

Problem

We needed to define how conversations should behave across:

Surfaces within Drive

File list → Overlay viewer → Standalone tab

Cross-product journeys

e.g. Gmail → Drive → Docs → Sheets

Two key questions emerged:

  • What should happen to a conversation when users move between surfaces or products?
  • When users return to a previous app, should the conversation continue, reset, or branch?
Problem visualization
Hypothesis

Users want conversations to persist across Workspace to support seamless, cross-product workflows—but expectations around returning to previous contexts were unclear.

Before

Panel behavior before

After

Panel behavior after

Exploration

Conversation Models

I defined three models to evaluate:

1. Always Mirror

A single conversation persists and mirrors across all apps and surfaces—following the user like an assistant.

2. Mirror When Contextually Relevant

Conversations persist and mirror only when they are part of a clear user journey, such as interacting with a file or entity tied to the task.

  • Opening a new tab without context → Zero state
  • Returning to an existing tab → Previous conversation is preserved
  • Contextual actions (e.g. clicking a file, chip, or entity) → Continue the same conversation

3. Always Branch

Each new surface or app creates a new branch of the conversation, even if related.

Conversation models exploration

Evaluation

We tested these models through two cross-product workflows:

  • Track receipts: Gmail → Drive → Sheets → Gmail
  • Analyze a sales meeting: Gmail → Drive → Docs → Gmail

Password: 123

Always Mirror won—decisively.

We established a clear product direction:

Conversations should persist and mirror across all Workspace products and surfaces, enabling a seamless, assistant-like experience.

Key Findings

1. Strong Mental Model: Gemini as an Assistant

Users consistently viewed Gemini as a persistent assistant, not a tool tied to a single surface.

2. Preference for Continuity

Participants overwhelmingly preferred a single, continuous conversation that follows them across apps.

3. "Unexpected Delight" Effect

Even users who initially assumed they would prefer separation changed their minds after experiencing continuity.

Design Considerations

The side panel should:

  • Clearly start a new conversation at any time
  • Seamlessly access conversation history across apps
Panel design

Maintain consistent open-state behavior across surfaces

Open panels behavior

Persistent Gemini conversations across Workspace products

Cross-product Gemini conversation flow

Outcome & Next Steps

Initial Launch

Drive introduced persistent conversations within the file viewer, enabling continuity when users cycle through files. This established the foundation for a more assistant-like experience within a single product surface.

Next Steps

To extend this model across Workspace, we identified a key dependency:

Cross-product conversation history

Supporting shared conversation history across products (e.g. Gmail, Drive, Docs, Sheets) has now been prioritized as a direct result of this work.

Once this foundation is in place, we will:

  • Introduce persistent conversations across Workspace
  • Enable seamless, cross-product workflows with a single continuous conversation
  • Fully realize Gemini as a system-level assistant that follows users across their work