Grounding
Tying a model's answer to specific sources, such as documents or search results, rather than its memory alone.
A grounded answer is based on material supplied at the time of the question, such as a company's documents, a web search or a database, and ideally shows where each claim came from. Grounding reduces hallucination, keeps answers current past the knowledge cutoff and lets people check the work. Retrieval-augmented generation is the most common way to ground a model. Google, OpenAI and others offer grounding with web search as a feature of their APIs. A model can still misread or misquote its sources.