Latency is a copywriter

Last year we shipped a voice interface that was, by most measures, fast. Response time under two seconds. Users hated it.

Not because it was wrong. Because it felt dismissive.

What we got wrong

Two seconds is objectively fast for an AI response. But in a conversation — a real, human conversation — two seconds of silence after a question feels like the other person is annoyed at you.

We'd been measuring latency as an engineering metric. It turns out it's also a communication one. How long something takes tells you how hard the system is thinking about you.

The fix

We added a small indicator that appeared within 400ms of a query. Not a spinner — just a subtle visual beat. Something that said: I heard you. I'm thinking.

Satisfaction scores went up. Perceived accuracy went up, even though the answers were identical.

The latency hadn't changed. The communication had.

The broader point

Every pause, every delay, every transition in a UI is making a statement. "This is fast" is a statement. "This is worth waiting for" is a different statement.

Voice UX makes this especially legible because there's nothing else happening. No animation, no loading state, no interface to read. Just time. And how you use that time is the interface.