# Grounding

> Grounding is the practice of tying an AI model's output to specific, verifiable sources, such as retrieved documents, database records or tool results, rather than relying on what the model learned in training. A grounded answer can point to where each claim came from. Retrieval-augmented generation is the most common grounding technique, often combined with required citations and source checks.

## Why it matters for your business

Grounding is what makes AI answers auditable. If an assistant says a contract auto-renews, a grounded system can show the clause; an ungrounded one just asserts it. For business use, grounding reduces hallucinations, keeps answers consistent with current policies, and lets reviewers check work quickly instead of researching it again. It also clarifies what went wrong when an answer is wrong: the source was outdated, the retrieval missed, or the model misread the source. That makes problems fixable rather than mysterious.

## Example

A nonprofit's grant writing assistant drafts proposal sections only from the organization's program reports, budgets and past funded proposals. Each paragraph carries footnote references to the source document and page. Before submission, the development director checks the footnoted figures, such as the number of families served, against the original reports instead of trusting the draft.

Related terms: https://www.lumeor.studio/glossary/rag-retrieval-augmented-generation/, https://www.lumeor.studio/glossary/hallucination/, https://www.lumeor.studio/glossary/knowledge-base/, https://www.lumeor.studio/glossary/evals-ai-evaluation/

---

Source: https://www.lumeor.studio/glossary/grounding/