mobile-adaptation

Mobile Adaptation

R — Original Text (Reading)

Claude Mobile iOS sets response tiers based on screen size: phones display 6-8 sentences at a time, simple questions get 1-2 sentences, how-to guides get short lists, substantive questions get 2-3 paragraphs, and complex questions do not exceed 2 screens. Integrates mobile-native tools (calendar, reminders, location, charts). Claude for Word prohibits pipe-delimited Markdown tables (the task pane is too narrow). Gemini implements mobile-specific output compression. Core patterns: screen-size-responsive tiering, mobile-native tool integration, format restrictions, answer-first strategy.

I — Methodology Skeleton (Interpretation)

1. Screen-size-aware tiering: Divide responses into 4 tiers based on the target device's screen capacity, with each tier having a clear length limit (number of sentences, paragraphs, or screens).
2. Answer-first strategy: Mobile users have fragmented attention, so the response structure must be "conclusion first, details after," with no introductory preamble.
3. Format restriction checklist: Disable specific formats in narrow-screen scenarios—pipe tables, deeply nested lists, wide code blocks, long block quotes.
4. Mobile-native tool integration: Leverage capabilities unique to mobile devices (calendar, reminders, geolocation, local time, chart display) to enhance interaction.
5. Scan-friendly structure: Use formats such as short lists, bolded keywords, and section headings so users can locate the core information within 3-5 seconds.

A1 — Case Analysis (Past Application)

Case: Claude Mobile iOS's four-tier response tiering

• Problem: A mobile screen can only display 6-8 sentences at a time, and overly long responses require extensive scrolling, severely affecting information acquisition efficiency in mobile scenarios.
• Use of the design pattern: Claude Mobile iOS divides responses into four tiers and sets strict length constraints—simple questions are answered directly in 1-2 sentences, how-to guides use the shortest possible lists, substantive questions get 2-3 paragraphs, and complex questions do not exceed 2 screens. All tiers follow the principle of "give the answer first, no preamble."
• Conclusion: Quantitative tiering based on physical screen constraints is far more effective than vague "keep it as short as possible" instructions, providing the model with an executable length standard.

Case: Claude for Word's table format prohibition

• Problem: The task pane of the Word add-in is extremely narrow (about 300-400px), and pipe-delimited Markdown tables overflow or wrap chaotically.
• Use of the design pattern: Claude for Word explicitly prohibits pipe-delimited Markdown tables in chat ("No pipe-delimited markdown tables in chat"), using structured lists or natural language descriptions instead.
• Conclusion: Format restrictions need to be specific to particular Markdown syntax elements; generalized "mind the format" instructions cannot precisely solve narrow-screen adaptation problems.

A2 — Trigger Scenarios (Future Trigger) ★

In what situations do users need this?

1. Designing system prompts for AI assistants embedded in mobile apps
2. Optimize existing desktop system prompts to adapt to mobile
3. Building cross-platform AI products requires differentiated output for different screen sizes
4. Develop an AI assistant integrated with mobile native features (calendar, location)

Language signals

• "mobile users"
• "displayed on a phone screen"
• "small screen adaptation"
• "needs calendar/reminder/location integration"
• "AI assistant within an app"

Distinction from adjacent skills

• Difference from voice-optimization : voice optimization focuses on the auditory channel, while mobile adaptation focuses on the physical constraints of the visual channel; however, both share the philosophy of conciseness first
• Difference from citation-system : citations require special presentation on mobile (such as simplified markers, collapsed citations), but mobile adaptation does not involve the design of citation formats itself

E — Executable steps (Execution)

1. Step 1: Define a screen response tier table - Completion criteria: Based on the target device's screen capacity, define a 4-tier response strategy (simple/operational/medium/complex), specifying for each tier the maximum number of sentences, paragraphs, or screens, with concrete examples.
2. Step 2: Write a format restriction checklist - Completion criteria: List the format types prohibited on mobile (pipe tables, deeply nested lists, code lines exceeding 60 characters, etc.), and provide alternatives for each prohibited format (table→structured list, nested list→flat enumeration).
3. Step 3: Design an answer-first output structure - Completion criteria: Declare the "conclusion first" principle in the system prompt, specify the response structure as: direct answer → key details → optional expansion, and prohibit introductory preamble.
4. Step 4: Plan mobile native tool integration points - Completion criteria: List the callable mobile native capabilities (calendar creation, reminder setting, location query, time retrieval), and define trigger conditions and invocation formats for each capability.
5. Step 5: Add scan-friendly format specifications - Completion criteria: Specify format enhancement rules for mobile output—bold key information, keep list items to no more than one line, separate paragraphs with blank lines, and use emoji prefixes (if applicable) to improve visual scanning efficiency.

B — Boundary (Boundary) ★

Do not use in the following cases

• Desktop-first system prompt design, where screen space is not the main constraint
• Mobile UI/UX design (belongs to front-end development, not the prompt layer)
• Pure voice interaction scenarios (no screen display, should use voice-optimization)
• Backend API design (unrelated to the output presentation layer)

Common failure modes

• Lack of quantitative standards: Only saying "keep it as short as possible" without giving specific sentence or screen limits, so the model cannot precisely control output length
• One-size-fits-all compression: Compressing all questions into one or two sentences, causing information loss for complex questions; should be handled by complexity tier
• Ignoring native capabilities: Only optimizing text output but not leveraging capabilities unique to mobile devices such as calendar, location, and reminders, missing opportunities to enhance interaction
• Format restrictions too general: Saying "pay attention to mobile formatting" without specifically pointing out prohibited syntax such as pipe tables, so the model may still output incompatible formats