Discussion starter: an entity SEO question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- Entity SEO
Have Google’s Natural Language API entity salience scores helped you optimise pages, or pointed you in the wrong direction?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
Entity salience scores from Google’s Natural Language API are useful as a sanity check and misleading as a target. The API is a Google Cloud product that reports which entities a text mentions and how central each appears to be. It is not the system Google Search uses to rank pages, and Google has never said salience scores relate to rankings. Used lightly, it can tell you whether a page is clearly about what you think it is about, and whether an ambiguous term is being read as the wrong thing. Used as an optimisation target, it pushes writers to repeat the main entity and remove anything else, which makes pages worse for readers without any evidence that it helps them rank.
What the API is good for
- Checking focus. If a page about a specific product shows a different entity as most salient, the page may be structured around the wrong thing, for example the brand’s history rather than the product.
- Spotting ambiguity. The API links entities to knowledge base entries where it can. If “Mercury” is linked to the planet on a page about the element, or a product name is linked to an unrelated entity, the text may be unclear to people and machines alike.
- Comparing coverage. Running competing pages through the API shows which related entities they discuss. The useful signal is coverage of relevant concepts, not the salience number.
Why salience is a poor target
Problem Why it matters Different system The Cloud API is not Google Search’s ranking system; its scores are not ranking signals Sensitive to position and repetition Moving a mention to the first sentence or repeating it raises salience without improving the page Rewards narrowness Removing related concepts raises the main entity’s share but makes the page less useful No documented link to outcomes No public evidence that raising salience improves rankings or AI citations How I would use it in practice
- Run the page and the top-ranking competing pages through the API.
- Check that your page’s most salient entity is the one it should be about. If not, read the page: is the topic clear in the title, heading and opening paragraph?
- Check entity links for ambiguity, and clarify wording where the API links to the wrong thing.
- List related entities that competing pages cover and yours does not. Decide, as an editor, whether covering them would help readers.
- Stop there. Do not iterate on wording to push the salience score higher.
Hypothetical example:
A content team rewrites product guides to maximise the salience of each product name, repeating it in every paragraph and removing sections about related accessories and compatible systems. Salience scores rise across the board. Over the following months, rankings for the main product queries do not change, while rankings for longer queries about compatibility and accessories fall, because that content is gone. The team restores the removed sections and uses the API only to catch pages where the main topic is genuinely unclear.
What matters more than salience
Pages rank and get cited when they answer the query well: the right information, clearly stated, with the related concepts a reader needs. Titles, headings and the first paragraph should make the topic obvious. Related entities should appear because they are relevant, not because a tool expects them. Structured data and internal links can help clarify what the page is about and how it relates to other pages on the site. All of these improve understanding for people and machines without chasing a score from a different system.
Watch for this:
Many content tools present entity or salience scores as optimisation targets with a “recommended” level. Those recommendations are derived from correlation with pages that already rank, not from how Google ranks. Treat them as hints for editorial review, and never trade clarity for a higher score.
Better tools for the same questions
Most of what teams hope to learn from salience scores can be learned more directly. To see whether a page is about the right thing, read its title, heading and opening paragraph as a stranger would. To see which related topics matter, look at the questions searchers ask around the query: related searches, the questions shown in results, your own site search and support tickets. To see how Google understands the page, check which queries it actually receives impressions for in Search Console; a page that gets impressions for unrelated queries is being read differently from how you intended. To check how Google treats the entity itself, the Knowledge Graph Search API shows whether a brand or concept is recognised. Each of these is closer to the systems and people that matter than a salience score from a general-purpose language API.
A note on AI answers
AI systems that retrieve and summarise content need to understand what a passage is about, and ambiguous text can lead to misattribution or misunderstanding. That is a reason to write clearly and name things precisely, which the API can help check. It is not a reason to believe that the API’s salience score predicts citation. No public evidence connects the two.
Use the Natural Language API to check topic focus and catch ambiguous entity links, compare which related concepts good pages cover, and leave the salience number alone. Clarity and usefulness are the targets; the score is at best a symptom.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.