VectorCiteVectorCite
All terms
metrics

Groundedness

How tightly a generated answer's claims map to retrieved source content — high groundedness = low hallucination risk.

Definition

Groundedness measures the degree to which every claim in an AI-generated answer can be traced to a retrieved source passage. High groundedness: every fact in the answer maps to a citation. Low groundedness: the model fills gaps with training-data knowledge or interpolation. Engines like Perplexity score groundedness internally and prefer to cite sources that contribute to high-groundedness answers.

Example

Answer with 4 facts and 4 citations, each citation supporting one fact = high groundedness. Answer with 4 facts and 1 citation = low groundedness.

How to optimize

Make your content easy to extract one fact at a time: short sentences, numbered facts, clear attribution to sources you cite. Avoid generic claims without backing.

Related terms

See groundedness in your audit.

Run a free 30-second audit and see where your page sits on this signal — with the concrete fix already generated.

Run free audit