Definition
Embeddings are dense numerical vectors (typically 384, 768, or 1536 dimensions) that represent the semantic meaning of text. Two pieces of text with similar meaning have vectors close in cosine distance. Embeddings are the substrate of dense retrieval: a query embedding is compared to document embeddings, and the closest documents are retrieved. Modern engines use bge-small, voyage-3, or proprietary embedders.
Text 'best CRM for agencies' embeds to a 1536-dim vector. Text 'top CRM solutions for marketing firms' embeds to a vector with cosine similarity ~0.93, so the engine treats them as nearly synonymous.
How to optimize
Use the actual buyer vocabulary on the page: embeddings reward semantic alignment, not just keyword presence. If buyers say 'CRM for agencies', use that phrase even if the SEO copywriter prefers 'agency-focused customer relationship management software'.