engines
Agentic search
When an agent decomposes a task into multiple search steps and acts on the results — the next generation past one-shot answer engines.
Definition
Agentic search is the pattern where an AI agent (not a one-shot answer engine) makes a sequence of search calls to accomplish a task. The agent plans → searches → reads → revises plan → searches more → synthesizes. Examples: ChatGPT Operator, Claude with computer use, Google's Project Mariner, Perplexity Pro Search. Citations may come from a chain of 10+ pages, not 5.
Example
User asks 'plan a 3-day trip to Tokyo under $1500'. The agent fetches hotels, fetches flights, fetches activities, iterates on the budget — citations span dozens of pages.How to optimize
Agents read deeper than one-shot engines. Cover subtopics exhaustively. Be the canonical resource for sub-questions adjacent to your main topic. Internal linking matters more — agents follow links to refine.