Search engines read documents as sets of entities rather than bags of words. People, places, organisations, and products are recognised and linked to known identities. Entity salience is the score that says which of those entities the document is actually about.
Frequency is a weak proxy for it. A recipe that names sugar eleven times is not a document about sugar. Salience is driven by position and context: entities in the title, the opening paragraph, and the headings weigh far more than ones buried in a list, and the surrounding language decides how the entity is interpreted. Google's Natural Language API returns a salience score per entity, which makes the concept directly observable on your own pages.
For content work this reframes optimisation. Instead of pushing a phrase to a target density, the question becomes whether the intended subject is the most salient entity in the document, and whether the entities that co-occur with it match the ones that appear on pages already ranking.
Structured data helps by naming entities explicitly, and clear introductions do more work than repetition anywhere else on the page.