Nielsen Norman Group research found that users treat site-specific AI chatbots like search bars. Nine participants with varying digital literacy levels tested 8 different chatbots, typing minimal queries and expecting fast, scannable responses. A longer answer was not a better answer.
The core finding cuts against a common design assumption: conversational, friendly AI does not build relationships on task-oriented sites. Users arrive with a question, not an appetite for dialogue. They typed as little as possible and wanted to extract information quickly, not engage with a persona.
The full article details how query patterns, response format, and scannability interact across 8 real chatbot implementations. If you are designing or evaluating a site chatbot, the specific failure modes documented here are worth reading before your next design review.
[READ ORIGINAL →]