Grounding.

The process of connecting an LLM's outputs to verified external information sources to improve accuracy and reduce hallucinations.

Grounding refers to anchoring an LLM's responses to external, verifiable information rather than relying solely on patterns learned during training. A grounded model can access current web pages, documents, databases, or tool outputs and incorporate that information directly into its response.

The most common grounding technique is Retrieval-Augmented Generation (RAG), where relevant documents are fetched from a knowledge base and inserted into the model's context. Other grounding approaches include web browsing (Perplexity's core feature), code execution (allowing models to verify computational claims), and database queries.

Grounding is essential for production AI applications because LLMs have a training cutoff date and cannot inherently access real-time information. A grounded system that knows when it doesn't know something — and retrieves the answer rather than confabulating — is dramatically more reliable than an ungrounded one.