As the practice of organisational development (OD) bifurcates into the traditional form of diagnostic OD and the emerging form of dialogic OD (Bushe & Marshak, 2009) it is especially important to obtain a better theoretical understanding of dialogic OD. This need is particularly true of Appreciative Inquiry (AI) as the most prominent form of dialogic OD. The purpose of the research was to build on current theory underlying AI. The specific research question addressed was: In the course of an AI intervention aimed at transformational organisational change, how do transitioning individual employees feel and make sense for themselves? The research adopted a multiple-case study design with a predominantly qualitative methodology. A form of theory-driven evaluation known as realist evaluation (Pawson & Tilley, 1997)—comprising of context-mechanism-outcome pattern configurations—was applied. The purposively selected sample consisted of six employees who had not transitioned well following an earlier diagnostic OD intervention with a comparable change objective. Data were collected longitudinally using a combination of diary prompts, direct observation, and interviews. The main findings indicate that: (a) three conceptually independent types of cognitive outcome patterns can occur under AI; (b) during-AI contexts of positivity-orientated activities and a safe environment each predispose transition towards particular types of cognitive outcome patterns; (c) certain pre-AI contexts, such as dogmatism at the level of the individual, influence the degree of transition; and (d) certain reflective mechanisms link particular contexts and affective outcome patterns. The findings of the study build on, and are partly presented in the format of Bushe’s (2013b) facets of generativity model and an analytic framework is presented which offers a “way of seeing” transition under an AI intervention. By studying employee transition under AI in a well-specified research design, using clearly defined and well operationalised constructs, the research contributes theory which is substantially more comprehensive than previously available and from which testable propositions can be developed. It thereby overcomes concerns of authors such as Golembiewski (2000) and Bushe (2007) regarding the incompatibility of AI—with its basis in social constructionism—and “rigorous” research.