Definition
A vector embedding is a numerical representation of the meaning of a piece of text (a word, sentence, paragraph or entire page) in the form of a long series of numbers — a vector with hundreds to thousands of dimensions. Texts with similar meaning get vectors that lie close together; 'cheap running shoes' and 'affordable running trainers' sit right next to each other in that space, even though they share no words.
Embeddings are produced by a language model trained to capture meaning. Search engines and AI engines use them to compare queries and passages: the question becomes a vector, every candidate passage is already a vector, and the passages with the smallest distance (usually cosine similarity) are retrieved. This is the mechanism behind semantic search and the retrieval step in RAG systems such as ChatGPT search, Perplexity and AI Mode.
For SEO and GEO this has concrete consequences. Retrieval works at passage level: a page is split into chunks of a few paragraphs, and each chunk gets its own embedding. A paragraph that fully answers a question on its own sits close to that question in vector space and gets retrieved; a paragraph that only makes sense with the rest of the page sits further away. Self-contained, complete passages therefore beat continuous argumentative text.
Topic coverage is measurable in embeddings too. A page that names the question, the synonyms, the related concepts and the entities occupies a larger and more precise part of the meaning space around the subject. Repeating one keyword barely moves the vector; adding a missing sub-question does. That is why substantive completeness and concrete examples pay off more than keyword density.
Marketers do not need to compute embeddings to benefit from them, but it helps to think in this model: which questions sit close to my subject, which passages answer them on their own, and which are missing? Traze uses embeddings to cluster search and AI questions, to determine which passages of a page already sit close to a question and where the gaps are, and to check that every written section is quotable on its own.