Designing Conversations That Hold: A Practical Guide for Conversation Designers in the LLM Era

Most conversational AI still breaks down after two or three turns. This guide explains why: not a language processing problem, but a failure to design for conversational structure and common ground, the two things that keep a human engaged with a system past the first exchange.

You'll walk through the four foundations (structure, common ground, human-centered design, and repair) with concrete before/after examples at every step, not just theory. Then it moves into practical ground: five specific shifts in how conversation designers need to work now that responses are generated live by an LLM instead of written in advance, from specifying behavior instead of scripting turns, to treating repair as a confidence decision instead of a phrase catalog.

Use it as a working reference for design reviews, not a one-time read.