Prompts For Everyone logo
PFE
Back to Glossary
Concept

Grounding

Connecting AI responses to verifiable sources of truth so it doesn't make things up.

Share

Definition

Grounding is the practice of connecting an AI model's responses to verifiable, factual sources of information. An "grounded" AI response is one that's based on real data — documents, databases, verified facts — rather than the model's training data alone. Grounding is one of the primary defenses against hallucination.

When an AI is grounded in your company's documentation, it answers based on what's actually in those documents rather than making up plausible-sounding information. RAG is the most common grounding technique, but grounding also includes practices like instructing the AI to cite sources, cross-reference claims, and explicitly state when it's uncertain.

Examples

1

A legal AI that always cites the specific statute or case law it's referencing in its answers

2

A product support AI grounded in your knowledge base that says "Based on our documentation..." before answering

Related Terms

RAG (Retrieval-Augmented Generation)HallucinationGuardrails

Frequently Asked Questions

How do I ground my AI in my company's data?
The most common approach is RAG: store your documents in a vector database, retrieve relevant sections when a user asks a question, and include them in the AI's prompt. This ensures responses are based on your actual data.
Is grounding the same as RAG?
RAG is the most common grounding technique, but grounding is the broader concept. You can also ground AI through prompt instructions ("Only answer based on the provided documents"), structured data access, and output verification.

Build prompts using this concept

Explore our prompt library and put grounding into practice with ready-to-use templates.

Build prompts using this concept