Generative Models in Creative Work: How LLM and Multimodal AI are Transforming A/B Testing

Kuanysh Kemeshova

Abstract: The article examines how, under the influence of generative artificial intelligence, primarily large language models and multimodal models, the methodology of A/B testing of advertising creatives is being reconsidered and reconfigured. The introduction justifies the research relevance of the topic by pointing to the rapid expansion of the market for generative AI technologies and draws attention to the methodological limitations of the classical A/B approach that have created the preconditions for the current technological and conceptual shift. The aim of the study is to analyze the transformations in experimental procedures, optimization principles, and the structure of associated risks within the framework of A/B testing. The research strategy is based on a systematic review of scientific publications and a content analysis of a corpus of academic articles and industry analytical reports. It is demonstrated that the introduction of generative AI can fundamentally increase the scalability of producing diverse creative solutions, and that the use of multimodal models enables their integrated synthesis and multidimensional evaluation. Particular attention is paid to the emerging optimization paradigm based on reinforcement learning (RL) methods, in which business metrics such as CTR become the direct maximization objective. The institutionalization of a hypothesis-driven approach, which increases the interpretability of results and the operational efficiency of experiments, is discussed separately, and the risks of algorithmic bias, creative homogenization, and erosion of consumer trust are analyzed in detail. The conclusions drawn in the final part of the article indicate a qualitative paradigmatic shift from empirical trial-and-error to a systematically structured experimentation methodology grounded in AI and scientifically justified principles. The article is addressed to AI researchers, practicing marketers, and digital advertising specialists interested in understanding and practically implementing advanced approaches to the optimization of advertising creatives.

Keywords: generative AI, large language models (LLM), multimodal AI, A/B testing, creative optimization, reinforcement learning (RL), click-through rate (CTR), hypothesis-driven AI, algorithmic bias, digital advertising

How to Cite?: Kuanysh Kemeshova, "Generative Models in Creative Work: How LLM and Multimodal AI are Transforming A/B Testing", Volume 14 Issue 12, December 2025, International Journal of Science and Research (IJSR), Pages: 2448-2454, https://www.ijsr.net/getabstract.php?paperid=SR251206161057, DOI: https://dx.dx.doi.org/10.21275/SR251206161057

Download Citation: APA | MLA | BibTeX | EndNote | RefMan