Google AI Mode's composer teardown

What works

• Familiar Google search bar shape keeps intimidation low for billions of search users.
• Personalized greeting (“Hi Brian”) makes AI Mode feel conversational without a chat thread.
• AI Mode is a labeled pill beside mic and Lens: opt-in to generative search, not the default box state.

What we would push on

• AI Mode vs classic search distinction may confuse users who expect ten blue links.
• No visible starter prompts on first load. Discoverability depends on opening + or AI Mode.

Business strategy

Google wants generative search without breaking the search bar habit. An explicit AI Mode toggle lets billions of users opt in gradually instead of replacing classic search overnight.

Tradeoff

Takeaway

Anchor AI in the search bar users already know. Make the generative path an explicit toggle.

What works

• One + menu covers uploads and mode switches. The bar stays clean until opened.
• Create images and Canvas sit beside attachments as peer tools, not buried in settings.
• Fast vs Pro model picker lives in the same menu: speed vs depth without leaving compose.

What we would push on

• Menu labels reference Gemini 3 models on a google.com surface. Branding may blur AI Mode vs Gemini app.
• No one-line descriptions under Fast/Pro for first-time users.

Business strategy

Google bundles attach, creative tools, and model tier in one menu so AI Mode can grow into a full compose surface without turning google.com into a chat app.

Tradeoff

Takeaway

Funnel attach, generative tools, and model tier into one progressive disclosure menu.

What works

• Composer expands vertically to fit attachment preview. The file is visible before send.
• Chip shows filename on hover; X removes without clearing the whole prompt.
• AI Mode arrow activates once content is ready: clear send affordance separate from classic search.

Business strategy

Multimodal queries keep users in AI Mode instead of bouncing to Lens or Drive. Visible chips make file context explicit before a generative run.

Takeaway

Keep attachments inside the compose card with remove chips, not as mystery context.

What works

• Voice takes over the full screen. The mic is the focal point, not a tiny bar icon.
• Live transcription (“la la la…”) shows the system is listening before submit.
• Close X returns to compose without sending. The escape hatch is obvious.

What we would push on

• No hint that voice submits into AI Mode vs classic voice search.

Business strategy

Voice lowers friction for mobile search users and keeps them on google.com instead of switching apps. Full-screen capture signals a different job than typing.

Tradeoff

Takeaway

Dedicated voice surface with streaming text beats a pulsing mic icon alone.

What works

• Lens is one tap from the composer row. Visual search is peer to typing.
• Modal supports drag-drop, file upload, and paste-link: three ingest paths.
• Stays on google.com; greeting remains visible behind the sheet.

What we would push on

• Modal blocks the compose flow. Users may not know they return to AI Mode after Lens.

Business strategy

Lens is a proven Google capability. Wiring it into AI Mode turns visual queries into generative answers without a separate product switch.

Tradeoff

Takeaway

Wire Lens into the same input family as text: same bar, dedicated upload sheet.

What works

• Uploaded image stays in the search bar thumbnail. Visual query is persistent.
• AI Mode tab beside All and Visual matches ties generative summary to Lens context.
• AI Overview cites sources (5 sites chip). Multimodal input still gets provenance.

What we would push on

• Tab proliferation (AI Mode, All, Exact, Visual, About) may overwhelm casual Lens users.

Business strategy

Closing the loop from Lens to cited AI Overview keeps users in Google’s trust model. Visual search becomes a path into generative answers, not a dead-end results page.

Tradeoff

Takeaway

Close the loop from Lens upload to cited AI summary in one results chrome.

What works

• Canvas activates as a chip inside the composer. Mode is visible before typing.
• Placeholder shifts to “Create anything”. Signals generative doc/code intent.
• Contextual starters (weekend plans, housewarming, scavenger hunt) teach jobs without a tour.

What we would push on

• Canvas vs Create images both live in + menu. Chip labels must differentiate outcomes.

Business strategy

Canvas extends AI Mode beyond Q&A into deliverables. Starters lower the blank-page problem so users try generative docs without learning prompt patterns.

Tradeoff

Takeaway

Pair mode chips with job-specific starters so users know what Canvas is for.

What works

• Create chip (banana icon) signals image generation distinct from Canvas.
• Placeholder “Describe your image” sets expectation: not general Q&A.
• Starters show concrete formats: cross-section, infographic, map. Edit, don’t invent.

What we would push on

• Create and Canvas both add chips. Only one should be active; verify mutual exclusivity in UI.

Business strategy

Image generation is a high-engagement mode that competes with dedicated creative apps. Dedicated placeholder and starters raise completion without leaving google.com.

Tradeoff

Takeaway

Image mode needs its own placeholder, chip, and starters: not a generic Ask Google box.

What works

• Starter-style queries produce structured lists with headings, bullets, and venue links.
• Inline source chips (Visit Bellevue +2) and right sources panel mirror search trust patterns.
• Follow-up “Ask anything” bar persists at bottom. The compose loop continues after the answer.

What we would push on

• Heavy results UI may surprise users who expected a single paragraph from a starter tap.

Business strategy

Starters are onboarding into AI Mode answers with search-grade citations. Google keeps trust patterns from classic search while teaching what generative results look like.

Tradeoff

Takeaway

Contextual starters are onboarding into AI Mode answers. Show where compose ends and sourced output begins.

The pattern

• Personalized greeting, AI Mode toggle, + for attach/Create/Canvas/models, voice and Lens on the bar.
• Mode chips change placeholder and starters; answers return with citations and a persistent follow-up compose bar.

Where it varies

• AI Mode vs classic search: toggle is explicit but the mental model may still blur.
• Lens, voice, and text each use different surfaces before converging on results.
• Canvas and Create images both arm chips from the same + menu.
• Gemini model names appear on a google.com surface.

Business strategy

Google AI Mode is generative search on the homepage: familiar bar, explicit opt-in, search trust patterns in answers. The bet is billions of search users adopt AI without leaving google.com.

Tradeoffs

Takeaway

Search-native compose with an explicit AI Mode lane. Steal the toggle, mode chips with starters, and cited answers. Clarify Lens/voice return paths and Canvas vs Create differentiation.