KBT Score

The KBT score is a numerical value between 0 and 1 that measures how often a web source's extracted factual claims are supported — rather than contradicted — by the Knowledge Vault. A higher score indicates greater factual reliability.

What the Score Measures

Unlike domain authority or PageRank, which measure a source's position in the link graph, the KBT score measures content quality directly. The computation starts with fact extraction: an information-extraction pipeline reads pages on a given domain and pulls structured subject–predicate–object triples. Each triple is then verified against the Knowledge Vault.

The outcome for each fact is one of three labels:

• Supported — the knowledge base contains a matching triple with high confidence.
• Contradicted — the knowledge base contains a directly conflicting claim.
• Uncertain — the knowledge base has insufficient evidence to verify or refute the claim.

The KBT score aggregates these outcomes across all extracted facts for a source. Supported facts contribute positively; contradicted facts contribute negatively; uncertain facts are handled probabilistically using a Bayesian model.

The Probabilistic Model

Dong et al. formulated KBT as a probabilistic graphical model. It jointly estimates:

1. The probability that each extracted triple is correct (given source accuracy).
2. The trustworthiness score of the source (given the accuracy of its extracted triples).

This mutual dependency is resolved iteratively, similar to how PageRank iterates to convergence. The result is a stable per-source score that reflects both the volume and the accuracy of a source's factual output.

Score Range and Interpretation

In the original experiments on 2.8 billion facts across 119 million web pages, scores ranged across the full 0–1 interval, with high-quality encyclopaedic sources clustered near the top and tabloid or low-accuracy sources clustered near the bottom. The score is not an absolute threshold but a relative ranking signal — what matters is the source's position relative to competitors in its topic domain.

KBT Score vs. Domain Authority

Domain authority (DA), as defined by tools like Moz or Ahrefs, predicts ranking potential based on the quantity and quality of inbound links. It is purely exogenous. The KBT score is purely endogenous — it says nothing about who links to a source, only whether what the source says is accurate. For a structured comparison of the two approaches, see Knowledge-Based Trust vs. PageRank.

Practical Implications for Publishers

While Google does not expose KBT scores publicly, the principles translate directly into editorial best practice. Pages that make precise, verifiable factual claims on topics well-covered by knowledge bases (Wikipedia, Wikidata, academic databases) are structurally better positioned to receive high factual-accuracy signals. Vague claims, hedged predictions, or factual errors all reduce the ratio of supported-to-contradicted triples — and therefore the implicit KBT-type score a ranking system would assign.