← Back to Glossary
Grounding
Linking LLM responses directly to verified external facts, documents, or data sources to prevent unbacked fabrications.
Why it exists
Prevents hallucinations by requiring the model to answer using provided factual context rather than parametric memory.
How it works
RAG systems inject retrieved source snippets into the prompt and instruct the model to answer strictly using the provided documents.
Related terms
Where you'll use this
roadmap