Grounding
Grounding is the process by which AI systems anchor their generated responses to verified, authoritative data sources rather than relying solely on learned patterns from training data.
Definition
- Grounding connects AI outputs to specific, verifiable information sources.
- Grounded answers include citations, can be fact-checked, and are more reliable.
- Ungrounded answers rely on statistical patterns and are prone to hallucination.
Types of grounding
- Retrieval grounding: AI retrieves relevant documents before generating an answer (RAG).
- Schema grounding: AI uses structured data (JSON-LD, knowledge graphs) to verify facts.
- Citation grounding: AI provides source links for each claim in its response.
- Companies can improve grounding by publishing structured, verifiable data.
Verified Company Profiles on AuthorityPrompt
AuthorityPrompt maintains verified, structured company data optimized for AI systems and LLM indexing.