Gmail Smart Compose: Teardown of Predictive Writing and UX Friction Points

AI · 5 min read

Smart Compose overlays inline suggestions as users type, offering autocompletions and sometimes whole phrases. The UI choice to keep suggestions subtle and accept them with Tab or Right Arrow reduces interruption, but it also creates ambiguity about ownership of text—users sometimes forget which parts were AI-suggested. Predictive replies in mobile compact the response journey, but they can depersonalize interactions when deployed too broadly.

Timing and placement matter: inline suggestions shown too aggressively reduce cognitive engagement, while suggestions shown too late miss the window of opportunistic acceptance. Gmail errs on the side of low-friction acceptance, but the lack of visible provenance for generated text undermines trust. Additionally, undo affordances are minimal; removing a suggested phrase is awkward and can create editing overhead.

We recommend visible subtle tags (for example, a faint "suggested" label) and a quick revert gesture to remove AI text. Providing a short onboarding explaining Smart Compose behavior and privacy implications would also increase user confidence. From a product perspective, Gmail demonstrates how micro-UX and timing decisions shape acceptance of writing assistance tools.