Rethinking directiveness in AI coaching chatbots

Organizational coaching is well established as an effective intervention for professional and personal development. However, ongoing debate exists regarding the relative merits of directive versus non-directive coaching styles, with non-directive approaches widely accepted as desired practice in human-to-human coaching. Recent advances in generative artificial intelligence (Gen AI), alongside the widespread adoption of applications such as ChatGPT, have created the conditions for Gen AI-powered coaching chatbots to emerge as potential coaching alternatives, with initial studies showing promising results. The advent of these coaching chatbots raise many questions including to what extent they should mimic what human coaches do, including their level of directiveness. To investigate this knowlede gap, two Gen AI coaching chatbots, one directive and one non-directive, were developed using Heron’s intervention analysis model to operationalise distinct coaching styles. A between-subjects quasi-experiment (n = 158) measured and compared these AI coaching style preferences across four dimensions: technology adoption, working alliance, goal attainment, and personality traits. Contrary to established coaching conventions, results revealed a higher rating for the directive coaching style with respect to technology performance expectancy, working alliance, and goal attainment. Users exhibiting high levels of extraversion, conscientiousness, and openness to experience rated the directive approach higher. These findings raise questions about the prevailing non-directive coaching paradigm in AI coaching contexts and suggest that Gen AI coaching chatbots for workplace use should incorporate directive features tailored to coachee personality. The implications for coaching practice, AI coaching design, and future research are discussed.