Designing for conversation

Designing for conversation

Interfaces are getting better at sounding fluent. That does not mean they are getting better at being clear.

A response can be polite, well-formed, and impressive while still leaving someone unsure what just happened, what to do next, or whether they can trust it.

That gap is where most conversation-interface friction lives.

Where things slip

When people say an assistant feels off, they are rarely describing one major bug. The pause before feedback appears feels ambiguous. The tone shifts between turns. A suggestion appears before the system explains why.

Each moment seems minor in isolation, but together they shape trust. Trust is cumulative.

A useful frame

We use a simple rule: conversation UX should reduce interpretation work, not add it.

If users keep decoding intent instead of making progress, the product is asking too much. So we design conversation as structured interaction design: feedback timing, hierarchy, commitment states, reversibility, and voice consistency.

What changed in practice

Once we framed the work this way, decisions became clearer. We shortened action-to-acknowledgement time, made confidence visible, separated generated suggestions from system actions, and wrote explicit fallback language.

None of this is dramatic. That is the point. The work is less about making conversation feel alive and more about making it legible.

Closing thought

Fluency is pleasant. Clarity is what earns trust.