The Landscape of LLM-based Search Agents: A Survey
Abstract
Search is undergoing a structural shift from ranked retrieval and static retrieval-augmented generation (RAG) to agentic information seeking. Large language models (LLMs) now plan queries, browse pages, inspect evidence, maintain state, and synthesize answers or reports whose credibility depends on traceable external support. The resulting literature is fast-growing but difficult to compare. Web agents, search agents, agentic RAG systems, deep research agents, and multimodal browsing agents often reuse similar planners, retrievers, memories, verifiers, and writers, yet operate under different task regimes, evidence environments, output artifacts, tool budgets, and judging protocols. This survey argues that LLM-based search agents should be compared through a Search-Agent Comparison Contract that jointly specifies the task regime, evidence environment, evidence unit, control policy, output artifact, and evaluation contract. Using this lens, we delimit the field against traditional information retrieval, static RAG, generic LLM agents, and graphical user interface (GUI) agents. We then organize representative systems through comparison fields and executable workflow components, synthesize learning recipes from prompting and trajectory distillation to reinforcement learning, process rewards, verifier-guided search, test-time scaling, and multimodal orchestration, and consolidate benchmarks spanning BrowseComp-style browsing, GAIA/HLE-style assistant and frontier-reasoning…
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.