Gemini 3 and Gen UI in Google Search

Discussion of launching Gemini 3 in Search day one, enabling everything from complex reasoning and math to coding interactive simulations. The team emphasizes how this allows users to truly ask anything and get effortless information, with the model creating calculators and widgets on-the-fly to solve problems.

• •Gemini 3 can handle complex reasoning, math, and code simulations to help solve user problems
• •Day one launch gives product teams immediate access to all capabilities for creative use cases
• •Enables Search's core mission: allowing users to ask anything and get effortless information
• •Teams can be more ambitious in product development with such a powerful foundation model

Deep dive into Generative UI where models control both text responses and page construction. The model learns to think like a designer, making decisions about graphs, layouts, typography, and spacing based on design system instructions. Designers now write system instructions that teach the model design rationale, enabling dynamic experiences that weren't possible months ago.

• •Models receive design language specs including color palettes, typography, and layout principles
• •Designers write system instructions teaching models to make design decisions (primary vs secondary information)
• •Shift from static, bespoke experiences to dynamic improv-style generation using component libraries
• •Instruction following and reasoning in Gemini 3 enables encoding design principles in natural language
• •Maintains consistency through design system constraints while giving models creative control

Exploration of how models create interactive simulations for complex concepts like lift on airplane wings or car engine mechanics. These visual, motion-based experiences with hover interactions provide understanding that static text or graphics cannot achieve, representing the future of AI as a visual medium.

• •Models code interactive simulations with sliders, vectors, and animations (e.g., lift demonstration with adjustable wing)
• •Motion and interactivity are game-changers for explaining complex mechanical systems
• •Visual QA process being developed to evaluate design quality of generated experiences
• •System instructions for visual quality assurance showing promising results
• •Represents shift toward AI taking more visual forms rather than text-only responses

Discussion of maintaining design taste and quality in generative experiences, managing latency through thoughtful UX design, and using 'thinking steps' to communicate model activity. The team focuses on designing the latency experience itself and identifying which components can be reused to reduce generation time.

• •Visual QA process ensures design taste and craftsmanship in generated experiences
• •Latency designed as part of UX - showing users what model is doing (calculating, drawing)
• •Thinking steps communicate model activity during generation delays
• •Engineering work to identify reusable components that don't need regeneration
• •Search team's core competency in latency reduction being applied to AI experiences

Introduction of Gemini 3 Flash for bringing frontier model capabilities to everyday search at scale. Flash enables sophisticated reasoning, problem-solving, and generative capabilities with 3x faster performance, making advanced AI accessible to many more users while maintaining quality.

• •Flash is 3x faster than Pro for many use cases while maintaining quality
• •Enables frontier model capabilities at the speed and availability needed for everyday use
• •Supports sophisticated reasoning, problem-solving, and widget generation at scale
• •Aligns with Google's mission to make information accessible to as many people as possible
• •Critical for running advanced AI features across Search's massive user base

Experiment bringing AI Mode to the bottom of AI Overviews on mobile, creating fluid transitions between search results and conversational AI. Users can expand AI Overviews to enter full conversational mode, eliminating the need to choose between different AI entry points.

• •Mobile experiment: expanding AI Overview opens full AI-first conversational experience
• •Follow-up box at bottom enables seamless transition to conversational mode
• •Removes user confusion about where to ask AI questions (main search vs AI mode)
• •Power users can go directly to AI mode for complex questions
• •Goal: make AI access as easy and fluid as possible regardless of entry point

Evolution of Search's personality from implicit characteristics (science-focused, quirky, celebratory) to explicit conversational persona. The team is designing how Search responds to personal questions, emotional queries, and casual interactions while maintaining its unique 'googliness' and nerdy, jovial spirit.

• •Search historically had persona through Easter eggs, doodles, and unexpected delights, but not through language
• •Now designing responses to personal/emotional queries (e.g., 'I'm feeling sad')
• •Collaborating with DeepMind and Gemini teams while developing unique Search flavor
• •Infusing 'googliness' - quirkiness, science celebration, nerdiness, jovial spirit
• •Product taste now includes thinking about conversational personality alongside technical features

Nano Banana and Nano Banana Pro enable sophisticated data visualization, particularly for sports analytics and real-time information. The model combines reasoning with Search's knowledge to create custom infographics, live stats comparisons, and visual representations that transform how users understand data.

• •Creates custom infographics for comparing player stats and sports data
• •Visualizes data in completely new ways beyond traditional tables and charts
• •Combines model reasoning with Search's real-time knowledge and tools
• •Enables live game recaps with graphs, stats, and visual summaries
• •Represents unison of powerful models with Search's comprehensive data access
• •Shopping integration: visual galleries with live prices, color switching via follow-up questions