Building Sequences of Ads Relying on Discourse Analysis

We propose a method for generating sequences of advertisements derived from product descriptions and targeting keywords. Each sequence functions as a narrative, guiding potential customers through a storytelling journey. The sequence begins by building brand awareness, then highlights key product features, and ultimately culminates in a persuasive call to action that encourages the viewer to purchase the product or engage a service. The structure mirrors a well-crafted discourse framework, with each stage contributing to a final “punchline” designed to prompt the desired user action. The discourse structure of the ad sequence is dynamically managed by a large language model (LLM) enhanced with discourse analysis data. This enables the LLM to generate not only coherent and compelling ad content but also a persuasive narrative flow. Additionally, the LLM manages targeting features, tailoring messages to specific audiences. By leveraging click data from previous, similar ad campaigns, the model refines sequences to improve both relevance and performance. This integration of storytelling, discourse analysis, and data-driven targeting enables the creation of highly personalized and adaptive ads that evolve over time to improve engagement and conversion rates. The approach applies advanced AI techniques to automate ad creation, providing a scalable solution for businesses seeking to optimize advertising strategies through data-informed, narrative-driven campaigns. Our experiments demonstrate substantial improvements in ad impressions and targeting when using discourse analysis–supported LLM-generated sequences.