Gmail’s AI Smart Compose v3: A Behavioral Design Teardown

AI · 6 min read

Smart Compose v3 expands beyond sentence completion to suggest tone, brevity, and next-sentence ideas. The UI keeps suggestions inline and dismissible, minimizing disruption while nudging users toward faster composition. This inline-first approach preserves the author’s flow but can subtly prime phrasing and formality, influencing communication norms over time.

Personalization is central: Smart Compose adapts to user vocabulary and commonly used templates, which lowers correction cost. Yet it also raises concerns about echoing biases and reducing rhetorical variety. Gmail mitigates risks by providing transparency controls, a visible 'AI suggested' badge, and easy correction interactions so users feel in command of edits.

Designers should treat predictive writing as a behavioral intervention that reshapes norms. Instrumentation needs to measure downstream effects—reply latency, politeness markers, and correction rates—to understand whether the assistant is truly enhancing clarity or subtly steering voice and style.