In this lesson you’ll cover:
- How a modern AI search system processes a query end to end, from the moment it’s typed to the synthesized response
- The distinction between queries, tasks, and objectives—and why the system optimizes for inferred intent rather than the literal query string
- Query decomposition: how complex queries are broken into multiple sub-tasks, each requiring different information from different sources
- The five-stage retrieval pipeline (candidate retrieval, initial filtering, re-ranking, claim extraction, synthesis selection) and what gets eliminated at each stage
- The multiple retrieval sources a system can draw from (web search, knowledge graphs, structured databases, training data, conversation context) and which ones content producers can actually influence
- The five filters that determine whether content survives: relevance, quality, recency, consistency, and extractability
- Why the evaluation unit shifts from the page to the individual claim, with concrete examples of what gets extracted versus discarded
- Practical implications for how to approach content optimization given this pipeline