Evaluating the quality of user interaction with AI interfaces: cognitive loads, UX metrics, and user loyalty

The article is devoted to the study of the influence of cognitive loads and UX metrics on the level of user loyalty when interacting with AI interfaces. The relevance is due to the widespread introduction of AI into digital products, which requires taking into account changes in the allocation of mental resources. The novelty of the work lies in an interdisciplinary analytical approach combining psychophysiological, behavioral and questionnaire assessment methods. The paper summarizes the results of more than 50 publications, including the theory of cognitive load, research on UX metrics, and mechanisms for displaying AI prediction uncertainty. Physiological sensors (EEG, oculometry), behavioral indicators (reaction time, number of errors) and questionnaires (NASA-TLX, SUS) were studied. Special attention is paid to the procedures for selecting metrics and algorithms for dynamic interface adaptation. The analysis of empirical and theoretical approaches has been carried out, the main hypotheses have been tested and recommendations for improving the user experience have been developed. The work aims to systematize measurement methods, establish the relationship between workload and loyalty, and offer practical recommendations. Methods of systematic review, mediation analysis and A/B testing are used for the solution. The conclusion describes the key findings and areas of application. The article will be useful for AI interface developers, HCI researchers, and UX specialists.