From general AI to custom AI: the effects of generative conversational AI’s cognitive and emotional conversational skills on user's guidance

Abstract

Purpose

Generative conversational artificial intelligence (AI) demonstrates powerful conversational skills for general tasks but requires customization for specific tasks. The quality of a custom generative conversational AI highly depends on users’ guidance, which has not been studied by previous research. This study uses social exchange theory to examine how generative conversational AI’s cognitive and emotional conversational skills affect users’ guidance through different types of user engagement, and how these effects are moderated by users’ relationship norm orientation.

Design/methodology/approach

Based on data collected from 589 actual users using a two-wave survey, this study employed partial least squares structural equation modeling to analyze the proposed hypotheses. Additional analyses were performed to test the robustness of our research model and results.

Findings

The results reveal that cognitive conversational skills (i.e. tailored and creative responses) positively affected cognitive and emotional engagement. However, understanding emotion influenced cognitive engagement but not emotional engagement, and empathic concern influenced emotional engagement but not cognitive engagement. In addition, cognitive and emotional engagement positively affected users’ guidance. Further, relationship norm orientation moderated some of these effects such that the impact of user engagement on user guidance was stronger for communal-oriented users than for exchange-oriented users.

Originality/value

First, drawing on social exchange theory, this study empirically examined the drivers of users’ guidance in the context of generative conversational AI, which may enrich the user guidance literature. Second, this study revealed the moderating role of relationship norm orientation in influencing the effect of user engagement on users’ guidance. The findings will deepen our understanding of users’ guidance. Third, the findings provide practical guidelines for designing generative conversational AI from a general AI to a custom AI.