Entity Salience Score

Google’s Natural Language API returns a salience field for every entity it extracts, defined in its own docs as information about the importance or centrality of that entity to the entire document. The number is not about sentiment or accuracy. It answers one question: is this page actually about this thing, or does the thing just get a mention?

Salience is weighted by where and how often an entity appears. A term used as the grammatical subject, introduced early, and referred to consistently tends to score higher than the same term buried in a closing sentence. Practitioners like Charlie Norledge (Impression) treat it as an on-page check, not a magic dial: lead with your target entity, make it the subject of sentences, keep capitalization consistent, and vary references (named, nominal, pronominal) instead of repeating the exact phrase.

Two cautions. First, the score you pull from the API is a proxy. It shows how one NLP model reads your text, a useful stand-in for how Google’s entity systems and LLMs parse it, but it is not a confirmed ranking factor. Second, a page can only be strongly salient for a few entities. Trying to be the main subject of everything dilutes all of them.

Quick check

Paste a draft into the API (or a tool that wraps it) and look at the top entities. If your primary topic is not near the top, with a score above roughly 0.2 to 0.3 on a focused page, the copy is probably ambiguous about what it covers.