Grounding

What is grounding?

Grounding connects an AI system's output or decision to relevant evidence, trusted data, rules, or observable real-world state. A grounded response should be supportable by information available to the system rather than relying only on plausible model generation.

Grounding can apply to answers, plans, tool choices, and claims about whether an action succeeded.

How grounding works

A system may ground an answer using retrieved documents, database records, tool results, user-provided facts, or verified environmental observations.

For example, a support agent should ground an order-status response in the current order record. It should not infer shipment from typical delivery times. For important claims, the response may include source or timestamp information.

Grounding versus RAG

[Retrieval-Augmented Generation](/glossary/retrieval-augmented-generation/) is one technique for grounding. It retrieves relevant information and places it into the model's context before generation.

Grounding is broader. An agent may ground a decision through an API response, calculation, policy rule, sensor reading, or user confirmation without using a document-retrieval pipeline.

RAG also does not guarantee grounding. The model can ignore retrieved evidence, misread it, or add unsupported claims.

Grounding versus factual correctness

A response can be grounded in a source that is outdated or wrong. Grounding shows the connection to evidence; it does not prove that the evidence is authoritative.

Systems should evaluate both evidence use and source quality. Conflicting sources need resolution rules, while missing evidence should produce uncertainty or clarification rather than invention.

Risks and controls

Untrusted content may contain malicious instructions. Evidence and instructions should remain separate, and retrieved documents should not override higher-priority agent rules.

Builders should preserve provenance, apply permissions during retrieval, verify time-sensitive state, and require confirmation for high-impact actions. Citations should point to the source that actually supports the claim.

Why it matters

Grounding reduces the gap between fluent output and reliable information. It makes agent behavior easier to verify and helps users understand what evidence supports a conclusion.

The practical goal is not simply to attach sources. It is to ensure that important claims and actions are based on relevant, current, authorized evidence and that unsupported uncertainty remains visible.