What is Grounding?
Grounding is a technique of providing AI models with factual, current data as context before generating a response. The goal: the model bases its response on real information, not on potentially outdated "memory" from training data.
Grounding and hallucinations
AI hallucinations (generating false but plausible-sounding information) are one of the biggest enterprise deployment problems. Grounding minimizes this risk: the model receives specific documents, database data, search results, and must base its response on provided materials.
Grounding methods
Common approaches include: RAG (searching relevant documents and including them in the prompt), function calling (model queries APIs during response generation), knowledge graphs (structural fact sources), and tool use (model uses calculators, databases, search engines).