Five qualities determine whether a site-specific AI chatbot earns user trust or gets abandoned: handoff willingness, flexibility, proactivity, emotional responsiveness, and transparency. Nielsen Norman Group lays out a design framework built around these dimensions, arguing that decisions made before user testing even begins will shape whether a chatbot succeeds or fails in deployment.
The stakes are real. Users are still building mental models of what site-specific chatbots are supposed to be, caught between expecting a human customer-support rep and expecting a frontier LLM. Most chatbots fail both comparisons and lose users fast. The first quality alone, handoff willingness, cuts to a core failure mode: chatbots that create friction when users want a human are chatbots users stop trusting.
The full article is worth reading for the specific behavioral benchmarks behind each quality, not just the labels. The framework is designed to evaluate chatbots that already exist, not only guide ones in development. If you manage or build a chatbot and have never stress-tested it against these five dimensions, you are guessing.
[READ ORIGINAL →]