dictionary
Grounding forces a generative model to base its outputs on provided context rather than its internal, pre-trained weights. A model is considered "grounded" when every claim it makes can be directly traced back and cited to an authoritative source document provided at query time.
The practice of tying an AI model’s responses directly to verifiable, factual data sources (like a company database or web search) to prevent hallucinations.
We help teams go from definitions to deployed workflows—safely and fast.
RAG (Retrieval-Augmented Generation) is the most common grounding technique. Before the LLM answers a question, a search system retrieves relevant PDF snippets or database rows. The LLM is then instructed: "Answer the user's question using ONLY the following retrieved snippets."
Generally, no. Fine-tuning updates the model’s internal weights, but the model can still hallucinate based on those new weights. Grounding specifically refers to providing explicit external evidence at the exact moment a response is generated.
Get a copy of “Grounding” and an AI readiness checklist in your inbox.