Definition

Grounding

What Grounding means

Grounding is the practice of tying a model's output to specific sources supplied at the moment of the request, rather than letting it answer from memory. Google's own parenthetical definition is "providing content from the Google Search index to the model at prompt time to improve factuality and relevancy."

More precisely

Google describes Grounding with Google Search as connecting the model "to real-time web content", returning citation annotations with start and end indices so an application can show which span came from which source. Cloudflare's Content Signals treat this as its own category, ai-input, defined as "inputting content into one or more AI models (e.g., retrieval augmented generation, grounding, or other real-time taking of content for generative AI search answers)", kept separate from search and from ai-train.

Why it matters

Citations come out of the grounding step, so whether a site can be cited depends on whether the grounding system can retrieve it. Google is careful about the size of the claim, describing grounding as reducing hallucinations rather than eliminating them.

Example

When an assistant answers with linked sources under the text, those links are its grounding set, not a bibliography written after the fact.

Sources

Where this came from

Definitions here follow the specification or documentation that defines the term, not a paraphrase of another glossary.

Keep reading

Related

Retrieval-augmented generation Google-Extended Learn