Mobile UX for AI Copilots: Small Screens, High Stakes

Desktop copilots get keyboards, wide layouts, and long sessions. Mobile copilots get one thumb, spotty networks, and constant context switches. On MTR commutes and quick client checks, people treat AI like messaging: fast send, fast feedback, low patience for walls of text.

Design input for one hand

Offer voice where privacy allows, plus chips for your top three tasks. Keep the composer sticky. Do not bury send behind menus. Autocomplete should not jump the cursor while someone is editing mid-sentence.

Stream with structure

Streaming helps perceived speed, but break content into short paragraphs and bullets as it arrives. Collapse citations behind taps so the main answer stays readable on a narrow screen.

If users cannot tell the app is still working within a second, they assume it froze.

Respect interruption

Persist threads and draft text across app switches. On network loss, offer retry without wiping context. Light haptics on send are enough for most power users.

Deep-link into screens you already trust

When the model suggests an action, open maps, calendars, product detail, or checkout in native UI. The copilot proposes. Your proven screens close the loop.