Definition
Entity salience is a measure of how central a given entity is to a piece of text. Google's Cloud Natural Language API reports it as a score from 0 to 1 for every entity it detects, which its documentation describes as 'the importance or centrality of that entity to the entire document text'. An article about Tesla's earnings might give Tesla a salience of 0.6 and Elon Musk 0.1, even though both are mentioned.
The idea comes from natural language processing research on telling the main subject of a document apart from entities that are only mentioned in passing. In Google's Natural Language API, each entity comes back with a type (PERSON, ORGANIZATION, CONSUMER_GOOD and others), its mentions in the text, its salience and, when matched, a Wikipedia URL and Knowledge Graph MID. Google does not publish how the score is computed or whether Search uses the same signal, so treat it as a diagnostic, not a ranking factor.
In GEO writing, 'entity salience' is also used more loosely for how strongly AI systems associate a brand with a topic across the web: whether an assistant names your company when someone asks about your category. That is a different measurement. It is better tracked with mention rate and AI share of voice per topic than with an NLP score, and it depends mostly on what other sites say. In an Ahrefs study of 75,000 brands (May 2025), branded web mentions had a 0.664 Spearman correlation with brand visibility in Google AI Overviews, the strongest factor measured.
Page-level salience matters for AI search because retrieval systems work with passages. If a product page talks mainly about the problem and refers to the brand as 'we' or 'our platform', a model that extracts that passage may not connect the solution to your name. Practical fixes:
- Name the entity explicitly, especially in the first sentences of a page and of each major section.
- Make the entity the grammatical subject of key claims ('Acme encrypts files at rest') rather than an aside.
- Keep one primary entity per page; split pages that try to cover three products equally.
- Use the same name form every time, not a mix of abbreviations and old product names.
Web-level association is built through third-party mentions that put the brand next to the topic: reviews, comparisons, analyst coverage, podcasts and community discussions. Topical depth on your own site supports it, as described under topical authority. Prompt tracking tools such as Promptwatch measure the result directly by recording how often a brand appears in answers to tracked prompts for each topic.
Salience is related to entity consensus and entity coherence but distinct from them: consensus asks whether sources agree on the facts, coherence asks whether the entity's own signals fit together, and salience asks whether the entity is the focus at all.
Examples of Entity Salience
- A cybersecurity vendor tests its 'Ransomware protection' page in Google's Natural Language API. 'Ransomware' scores 0.41 and the vendor's name 0.03, because the copy says 'our solution' throughout. After rewriting key sentences to use the brand name, the brand becomes the second most salient entity on the page.
- A comparison article covers eight project tools but gives one of them three times the coverage. The NLP salience scores reflect that, and AI answers that cite the article tend to lead with that tool.
- A meal-kit company tracks 60 prompts about 'healthy meal delivery' and is named in 9% of answers, while a competitor appears in 41%. Its own pages are well written, so the team shifts budget toward review coverage and creator partnerships.
- An agency splits a client's single 'Solutions' page, which described three products equally, into three product pages so each one has a clear primary entity.
