Automation of creative hypothesis testing in video marketing based on large language models and computer vision

Abstract

The relevance of the study is determined by the growing role of video marketing in the digital environment, the increasing complexity of audiovisual content structures, and the rising requirements for the validity of managerial decision-making, which limits the effectiveness of traditional approaches to testing creative solutions and actualizes the need for implementing automated analytical systems. The purpose of the study is defined as the development of an integrated methodological framework for automating the testing of creative hypotheses in video marketing based on large language models and computer vision in order to enhance the validity, efficiency, and scalability of data-driven decision-making processes. Methods. The study applies methods of analysis and synthesis to generalize theoretical approaches to the automation of creative hypothesis testing; systematization – to structure their components and functional characteristics; comparative analysis – to assess the capabilities of large language models and computer vision; and logical generalization – to develop methodological approaches and practical recommendations. Results. The essence of creative hypotheses as multidimensional analytical constructs integrating semantic, visual, and behavioral parameters of content has been investigated. It has been revealed that the use of multimodal technologies ensures a transition to continuous testing and adaptive optimization of video content. It has been proven that the integration of large language models and computer vision into a unified analytical system increases the accuracy of content effectiveness evaluation, reduces experimentation costs, and minimizes the time lag between content creation and analysis. It has been established that the effectiveness of automation is constrained by issues of model interpretability, data quality and consistency, as well as the limitations of performance evaluation systems. Conclusions. It has been substantiated that improving the effectiveness of automated testing is achieved through the formation of an integrated multimodal data infrastructure, the implementation of closed-loop analytical systems, and the combination of experimental and attribution-based evaluation approaches. The necessity of aligning algorithmic optimization with long-term creative and brand strategies has been determined. Prospects for further research are associated with the development of methods for quantitative evaluation of multimodal content effectiveness, improvement of model interpretability, and investigation of the scalability of automated systems under conditions of dynamic digital environments.