AI Input & Context
AI input interfaces determine how users communicate with AI — from the initial entry point to multi-modal attachments to submit controls. We analyzed input patterns across ChatGPT, Claude, Gemini, Perplexity, Copilot, GitHub Copilot, v0, Notion AI, Google Docs, Airtable, Retool, and Atlassian Rovo. The data covers 6 entry point types, 4 interface placements, and over 150 composition details. Product context drives the biggest design decisions: standalone chat apps use persistent bottom panels, document editors use invoked modals, and workspace tools offer dual interfaces.
Entry Points
7 of 12 systems (58%) use always-visible input requiring zero invocation
All 5 standalone systems plus v0 and Retool AI keep the input field permanently visible at the bottom of the screen. Users simply start typing with no trigger required. This is the dominant entry pattern in the study, appearing in ChatGPT, Claude, Gemini, Perplexity, and Microsoft Copilot alongside both specialized tools.
Make the input field always visible for standalone AI products. Zero-click access is the universal standard for systems where AI interaction is the primary use case.
Entry Points
Slash commands appear in 3 of 12 systems, all productivity tools
Notion AI, Airtable Omni, and Atlassian Rovo are the only systems using slash commands to invoke AI. All three are productivity tools where slash commands already exist for other actions like inserting tables or mentions. Zero standalone chat systems use slash commands because the entire interface is already the chat — slash menus constrain what users think AI can do by presenting a fixed menu of options, undermining the open-ended nature of chat.
Add slash commands for AI only if your product already uses them for other actions. For standalone chat, use example prompts instead — they demonstrate capability breadth without implying a limited set of options.
Entry Points
How integrated products handle the discoverability problem
4 of 5 integrated systems offer 3 or more ways to invoke AI: slash command, keyboard shortcut, and menu button. Notion AI, Google Docs AI, and Atlassian Rovo each provide three distinct entry points. This redundancy ensures both new users (who need visual buttons) and power users (who want keyboard shortcuts) can find and use AI features.
Provide at least three entry points when integrating AI into an existing product. Include a slash command, a keyboard shortcut, and a visible button to cover all user skill levels.
Entry Points
How 3 systems teach keyboard shortcuts for AI
Data, screenshots, and actionable recommendations. Unlock this and every Pro insight.
See plans →Interface Placement
100% of standalone chat systems use a persistent bottom panel
All 6 systems with chat-first interfaces place the input at the absolute bottom of the viewport: ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, and v0. Every one shares the same formula: fixed bottom position, multi-line expansion, always visible, and a send button on the right side. No chat system uses a modal overlay — modals force a dismiss-reopen cycle between messages that destroys conversational flow. The persistent panel mirrors messaging apps like iMessage and WhatsApp.
Use a persistent bottom panel for any chat-first AI product. Never use a modal for multi-turn conversation — modals break flow with their dismiss-reopen cycle. This pattern has 100% adoption because it matches deep muscle memory from messaging apps.
Interface Placement
Document editors choose modal overlays at 100%
All 3 document-focused systems use modal overlays for AI: Notion AI, Google Docs AI, and Airtable Omni. Notion shows a contextual popup at the cursor position, Google Docs centers a panel with a backdrop, and Airtable presents a full modal with mode selection. None use persistent bottom panels because those would compete with the primary document editing area.
Use modal overlays when AI is secondary to document editing. The temporary interruption focuses attention on the AI task, then returns users to their content.
Interface Placement
Why 3 systems run two AI interfaces simultaneously
Data, screenshots, and actionable recommendations. Unlock this and every Pro insight.
See plans →Context Awareness
Context transparency scales with context complexity across all 12 systems
Zero of 5 standalone systems show context indicators, because users understand that AI sees only the conversation. 4 of 5 integrated systems show explicit context: Atlassian Rovo displays three simultaneous indicators (context badge, sources panel, and filter settings), while Notion shows @-mentioned pages in the prompt. Airtable Omni is the only integrated system that fails to indicate what AI can access, leading to prompts referencing content AI cannot see. The pattern is clear: broader context access requires more transparency.
Match your context transparency level to the complexity of what AI can access. Conversation-only needs nothing. Multi-source workspace access needs explicit badges, source panels, and filter controls. Hiding context in multi-source products creates user confusion and failed queries.
Context Awareness
How integrated systems scope what AI can see
Data, screenshots, and actionable recommendations. Unlock this and every Pro insight.
See plans →Text & Formatting
75% of systems use plain text input with markdown-rendered output
9 of 12 systems accept plain text and render markdown in the output. This covers all 5 standalone systems, both specialized tools, and 2 integrated systems. Only Notion AI and Google Docs AI use native rich text, integrating directly with their host editor's formatting so AI sees and preserves bold, italic, and heading styles. Airtable Omni is the sole plain-text-only system. The split is clean: native rich text only makes sense when the product itself is a rich text editor.
Default to plain text input with markdown rendering for chat-style AI products. Use native rich text only when your product IS a document editor where formatting must be preserved end-to-end.
Attachments & Files
7 of 12 systems support file attachments, all standalone systems included
7 of 12 systems support file attachments through a combination of upload buttons (paperclip or plus icons), clipboard paste, and drag-and-drop. All 5 standalone systems offer multi-method file upload. Google Docs AI and Airtable Omni skip attachments entirely because they already have document or database context built into the interface.
Support at least three upload methods: button click (required), clipboard paste (power users), and drag-and-drop (desktop users). Skip attachments only if your AI already has the content context it needs.
Multi-modal Input
Voice input is a standalone feature at 80% adoption vs 20% integrated
4 of 5 standalone systems support full voice conversation: ChatGPT, Gemini, Perplexity, and Microsoft Copilot. Only 1 of 5 integrated systems (Google Docs AI) supports even basic voice input, and that is transcription-only with no voice response. Claude is the sole standalone system without voice. Zero specialized tools support it.
Add voice conversation to standalone AI products where users expect full capability. Skip voice for integrated tools where document and database workflows dominate.
Controls & Submit
Arrow up icon is the submit standard at 67%, used by 8 of 12 systems
The arrow up icon dominates submit button design: ChatGPT, Claude, Gemini, Perplexity, v0, Airtable Omni, Retool AI, and Atlassian Rovo all use it. Only Microsoft Copilot and Google Docs AI use text labels ('Send' and 'Insert'/'Generate'). GitHub Copilot and Notion AI use sparkle icons to signal AI-specific submit actions.
Use the arrow up icon for your submit button in chat-style AI products. It is compact, universally recognized from messaging apps, and requires no localization.
Controls & Submit
Enter means submit in 83% of systems, with document editors as the exception
Data, screenshots, and actionable recommendations. Unlock this and every Pro insight.
See plans →Controls & Submit
Input modifiers appear only in systems with broad context access (3 of 3)
Data, screenshots, and actionable recommendations. Unlock this and every Pro insight.
See plans →Enhancements
All 5 standalone systems show starter prompts in the empty state
Every standalone AI product displays example prompts in the empty state: ChatGPT shows rotating suggestions, Claude offers categorized chips (Write, Learn, Code), Gemini provides action-oriented starters (Create image, Write anything), Perplexity displays mode selector icons, and Microsoft Copilot shows two rows of prompt chips. A blank screen with only an input field creates anxiety and fails to teach users what AI can do. This is the one pattern with absolute 100% adoption across all standalone systems.
Show 3 to 4 clickable starter prompts in every empty state. Categorize them by task type to demonstrate capability breadth and reduce the 'what should I ask?' problem. Make them demonstrate different capabilities rather than listing similar tasks.
Enhancements
How systems let users choose their AI model or mode
Data, screenshots, and actionable recommendations. Unlock this and every Pro insight.
See plans →Enhancements
How 3 standalone systems let users build custom AI personas
Data, screenshots, and actionable recommendations. Unlock this and every Pro insight.
See plans →Enhancements
How Airtable scopes AI with four distinct modes
Data, screenshots, and actionable recommendations. Unlock this and every Pro insight.
See plans →