Grounding
Definition
The practice of anchoring an agent's outputs to verifiable, authoritative sources of information — retrieved at inference time rather than encoded in model weights. Grounding reduces hallucination by giving the model specific, current, and contextually relevant evidence to reason from. The primary grounding mechanism is retrieval-augmented generation (RAG): before responding, the agent retrieves relevant documents from a knowledge base or the live web and injects them into its context. Grounding can also be achieved through tool use (querying a database for current values), structured data access (reading from a live API), or verification steps (cross-checking generated claims against retrieved sources).
Builder Context
Grounding is the most reliable way to reduce hallucination in production agents. The key design decisions: (1) what to ground — ground everything time-sensitive, domain-specific, or high-stakes; don't ground general reasoning; (2) retrieval quality — your agent's accuracy ceiling is set by your retriever, not your model; invest in chunking strategy and embedding quality before upgrading model size; (3) citation discipline — require your agent to cite sources in its output and validate those citations programmatically, not just in evaluation. Common mistake: treating grounding as a retrieval problem. Retrieval is the mechanism — the real problem is knowing what to retrieve (query formulation) and what to do when retrieval fails (fallback behavior).