Abstract
A system orchestrates stateful, multi-chapter advertising narratives across different digital platforms within real-time bidding environments. The system separates the ad-serving process into an asynchronous control plane for computationally intensive planning and a synchronous data plane for low-latency auction execution. The asynchronous control plane uses a large language model (LLM) to generate a narrative directed acyclic graph (NDAG) that defines a story arc with branching logic based on potential user engagement. The system then compiles this narrative graph into a deterministic state transition matrix for execution at the network edge. To track user progression, the synchronous data plane uses a compact narrative state vector (CNSV) stored in distributed edge caches for fast retrieval during a live auction. By reading the CNSV and using the pre-compiled matrix, the system determines the next chapter in the narrative sequence and serves the corresponding ad creative without requiring real-time artificial intelligence (AI) inference. This delivers a coherent and adaptive brand story across multiple surfaces by synchronizing ad exposures into a progression that responds to user interactions.
Keywords: Stateful Narrative Orchestration Engine (SNOE), Narrative Directed Acyclic Graph (NDAG), Asynchronous Control Plane, Synchronous Data Plane, Compact Narrative State Vector (CNSV), Real-Time Bidding (RTB) Networks, LLM Sequential Reasoning, Engagement-responsive Story Branching, Cross-surface Ad Narrative Orchestration
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Yakar, Tamar; Labzovsky, Ilia; and N/A, "Cross-Surface Ad Narrative Orchestration Using LLM Sequential Reasoning with Engagement-Responsive Story Branching", Technical Disclosure Commons, (September 02, 2026)
https://www.tdcommons.org/dpubs_series/11541