Microcopy for AI Assistants: A/B Study Shows Tone Trumps Transparency

AI · 4 min read

Several SaaS startups integrated generative assistants but struggled with balancing candid model uncertainty against user experience. The research team ran A/B tests across onboarding, in-product help, and email-drafting assistants with variants: candid transparency (detailed confidence and caveats), concise empathetic tone, and a hybrid.

Surprisingly, the empathetic tone variant produced higher task completion and lower follow-up clarification requests, even though users reported slightly lower perceived transparency. Interviews revealed that users preferred clear, actionable next steps delivered in a friendly voice over verbose caveats that created decision paralysis.

Design leads at the participating startups used the study to craft guidelines: favor concise, empathetic microcopy for primary assistant responses and reserve detailed transparency for an optional 'Why this happened' panel. This pattern allowed teams to maintain trust while optimizing for usability and task completion.