A Novel Taxonomy of Contextual Connection and Linguistic Dependency Developed for User Engagement with a Healthcare AI Chatbot

In AI–human conversations, engagement emerges through exchanges between users and AI, where each prompt may relate contextually and linguistically to prior turns. This study introduces a user prompt–AI response–user prompt (UP-AIR-UP) triad approach to examine these relationships and proposes a two-level taxonomy. The first level captures contextual connections between prompts (topic shift, prompt-to-prompt, prompt-to-AI response, and integrative prompts), while the second level assesses linguistic dependency (dependent vs. independent). Using data from healthcare chatbot interactions (106 prompts), results show that only 28% of prompts reflected contextual connections to AI responses or integrative behavior, with most prompts being linguistically independent. This pattern may suggest a less interactive mode of conversation, with users often treating chatbots as search tools. Unlike subjective self-report or physiological measures, this approach offers a low-cost, objective way to evaluate user interaction and can support models of user engagement.