Generative AI and Search Engine Orchestration: A User-Controlled Mode Selection and Adaptive Orchestration Framework

The increasing availability of generative artificial intelligence (GenAI) systems and conventional search engines has created two major but often disconnected approaches to accessing and processing internet information. Generative AI systems support content creation, explanation, reasoning, summarization, and other language-intensive tasks, whereas conventional search engines provide access to indexed webpages, original sources, official websites, and navigable information resources. Although these capabilities increasingly overlap, users are still frequently required to determine which interface or tool should be used for a particular task. This separation creates additional interaction costs, complicates multi-stage information workflows, and may lead to inappropriate tool selection when user intentions are ambiguous or when a task requires several complementary capabilities. This paper proposes a user-controlled mode selection and adaptive orchestration framework within Generative AI and Search Engine Orchestration (GASEO). Unlike an exclusively automatic routing architecture, the proposed framework incorporates an explicit user-facing mode selection mechanism alongside automatic orchestration. Users may select Generative AI mode, Search Engine mode, Hybrid mode, or other available modes and capabilities introduced through an extensible mode registry. When the user selects a mode, the system treats that selection as an explicit execution preference or constraint rather than merely another weak signal in intent classification. When the user chooses automatic orchestration, the system determines the appropriate execution strategy based on inferred intent, task composition, information freshness, source requirements, capability availability, and routing uncertainty. The framework further supports user overrides, mode switching, controlled hybrid execution, and the preservation of original search sources and generated content as distinguishable output components. The central contribution is a conceptual architecture for combining user-directed execution, system-directed orchestration, and extensible execution modes within a unified internet task interface. The paper formalizes the relationship between user control and automatic routing, proposes a mode-selection policy, describes interaction and execution workflows, and develops a testable evaluation agenda. It also distinguishes explicit user preference from inferred intent and separates mode selection from the internal orchestration of tools within a selected mode. The proposed framework is presented as a research architecture rather than an empirically validated system. Its potential contribution lies in investigating whether a unified interface can reduce the need for users to understand the technical distinctions between internet information-processing capabilities while preserving their ability to determine how their requests are executed.