Skip to main content
Back to Glossary
Techniques

Grounding

Connecting AI responses to authoritative sources or real-world data to reduce hallucinations and improve accuracy.


Why Grounding Matters

Ungrounded AI makes things up. It generates plausible text without checking facts. Grounding forces the AI to reference actual information.

Ungrounded: "The company was founded in 2018." (might be wrong)

Grounded: "According to their website, the company was founded in 2019." (verifiable)

Grounding Methods

Document retrieval (RAG): Give the AI relevant documents to reference.

Web search: Let the AI search and cite web sources.

Database queries: Connect to authoritative data sources.

API access: Call services that provide verified information.

Benefits of Grounding

  • Fewer hallucinations
  • Verifiable claims with sources
  • More current information
  • Domain-specific accuracy

Limitations

Grounding helps but doesn't eliminate errors:

  • Retrieved sources might be wrong
  • AI might misinterpret sources
  • Sources might be outdated
  • AI might ignore grounding for some responses

Evaluating Grounded AI

When reviewing AI tools that claim grounding:

  • Does it actually cite sources?
  • Can you verify the citations?
  • How current are the sources?
  • What happens when sources conflict?

Good grounding is transparent. You should see where information comes from.

Related Terms

More in Techniques