Entity salience

Google’s Cloud Natural Language API returns a salience value in the [0, 1.0] range for every entity it extracts, and that number is the clearest public window into how a machine judges what a page is “about.” Google defines it as the importance or centrality of an entity to the entire document, not how often the word appears. An entity named once near the top of the page often outscores one repeated ten times deep in the body, because practitioners consistently observe that document position and the way mentions cluster around one subject matter more than raw counts.

Reading a salience response

Run a page through the API and you might see:

"coffee"  salience 0.73
"water"   salience 0.17
"mug"     salience 0.04

The 0.73 says coffee is the dominant subject; the low scores mark water and mug as supporting detail. If your target entity comes back low, the page is diluted.

The Cloud NL API is a proxy, not Google Search’s live ranking signal, but it models the same idea: modern retrieval leans on entities and their relationships (the Knowledge Graph, topic embeddings) far more than on raw keyword counts. Note that the API reads plain text, so it never sees your HTML tags; what it rewards is a clear, early, consistently referenced main entity. To raise salience, name and classify the main entity early, connect it to its attributes and related entities, and keep one dominant entity per URL instead of chasing several at once.