Entity density is the number of recognized named entities (companies, products, people, places, technologies, integrations, dates) per thousand words on a page. Pages with at least fifteen entities per thousand words are roughly five times more likely to get cited in Google AI Overviews than pages below that threshold. The number is so consistent across categories that it is now treated as a hard target, not a vague heuristic.
Why entity density matters to retrievers
Modern retrievers run named entity recognition (NER) over every chunk during indexing. Entities are stored alongside the chunk's vector, and they get extra weight at retrieval time when the query mentions them. A chunk that mentions Salesforce, HubSpot, and Pipedrive together gets scored higher for a CRM-comparison query than a chunk that says 'leading customer relationship management platform' three times. The retriever is asking: how concretely is this passage anchored in the same world the query is about?
What counts as an entity
Brands and product names count. Personal names count. Company integrations count. Specific technologies (React, PostgreSQL, GPT-4) count. Cities, countries, regulatory frameworks (GDPR, HIPAA), and recognized standards count. Specific dollar amounts and dates count when they are tied to a named thing — '$29 per seat' on its own is weak, '$29 per seat on the Starter plan' is strong. Generic descriptive phrases like 'leading,' 'enterprise-grade,' or 'industry-standard' do not count and actively dilute density.
How to raise density without ruining the page
Three moves. First, replace generic descriptors with specific named referents — instead of 'leading e-commerce platforms' write 'Shopify, BigCommerce, and Salesforce Commerce Cloud.' Second, add a comparison sentence to every claim that currently has none: '...unlike Salesforce, which charges per-user' raises both density and search relevance. Third, end each section with one or two factual lines that pack entities tightly: 'Used by Notion, Linear, and Vercel for production search.' Each of those is a citation moment.
Where the density sweet spot lives
The data flattens above roughly 25 entities per thousand words. You do not gain much from going further, and at very high densities the prose starts reading like a press release rather than an explainer. A target of 18 to 22 entities per thousand words on your highest-priority pages is a reasonable goal. Run the page through any NER tool to count, then prune phrases that are dead weight and add a small number of high-relevance entities where the page already wants them.
Pages that probably need work
Three page types are usually below threshold. Hero-and-feature landing pages full of marketing prose. Blog posts written for thought leadership rather than search. Help center articles whose entities are buried in step-by-step procedures. Each of these can usually pick up 5 to 10 entities per thousand words with a single editing pass that swaps abstract phrases for named ones. That is enough to materially shift Overview citation odds.