Gmail's AI Compose Teardown: Balancing Assist and Authorial Voice

AI · 5 min read

Gmail's AI writing assistants reduce friction for common messages, but the current integration often treats suggestions as temporary overlays rather than parts of the draft's narrative. Users frequently accept sentence-level suggestions without a clear way to request a different tone or degree of formality from the same interface.

The blending of AI-generated content with user input creates attribution ambiguity: should accepted suggestions be editable with explicit regeneration, or should the assistant behave like a co-author with versioning? Current undo/redo behavior works but doesn't support explaining why a suggestion was made — e.g., referencing past emails or calendar context.

Design solutions include a 'Tone & Context' quick bar to set constraints before composition, an explicit 'Regenerate' affordance visible at sentence and paragraph levels, and a lightweight provenance overlay that explains which inputs the model used. These moves would make AI assistance more predictable and respectful of authorial voice.