Glossary
Grounding
Anchoring AI outputs in verified source data to reduce hallucination.
Definition
Grounding refers to techniques that anchor LLM outputs in verified source data — typically via retrieval, cited references, or restricted context. A grounded response is one where the model's claims can be traced back to specific input documents rather than parametric memory.
Context
Grounding is the primary lever for producing enterprise-safe AI output. RAG is the most common grounding pattern; others include tool use for verified data lookup, and citation-required output formats. Ungrounded generative AI is inappropriate for most enterprise use cases where wrong-but-confident outputs create liability.
Related terms
Retrieval-Augmented Generation (RAG)Injecting retrieved documents into an LLM prompt to ground outputs.Read HallucinationWhen an LLM generates confident-sounding output that is factually incorrect.Read Large Language Model (LLM)A neural network trained on massive text corpora to predict next tokens.Read