Definition
Embedding
Also known as: Vector embedding, text embedding
An embedding is a fixed-length list of numbers (a vector) that a neural model produces to represent the meaning of a piece of text, so that texts with similar meaning end up close together in the vector space. Answer engines embed both the question and candidate passages, then retrieve the passages whose vectors are nearest to the question's.
Embedding models take text in and return a vector. OpenAI's text-embedding-3-small returns 1,536 numbers by default and text-embedding-3-large returns 3,072 (OpenAI documentation). Similarity is usually measured as cosine similarity: vectors pointing the same way score near 1, unrelated ones near 0. A retrieval system computes the vector for each chunk of a page once, stores it in a vector index, and at query time compares the question's vector against millions of stored ones.
Two properties of embeddings shape how you should write.
They capture meaning, not spelling. "How do I stop OpenAI from crawling my site" and a passage about GPTBot in robots.txt land close together even though they share almost no words. Synonyms and paraphrase are handled for you; what must still be present is the concept and the named entity.
They are computed per passage, in isolation. The model never sees the heading three sections up or the antecedent of a pronoun. A chunk that begins "It is also worth noting that this approach…" embeds as vague filler; one that begins "Answer-first content places the direct answer…" embeds as a strong match for the question it answers. Production retrievers usually pair embeddings with a keyword scorer such as BM25, so exact terms still count, but the vector side is where answer-first structure pays off.
Frequently asked questions
Do I need to understand embeddings to do AEO?
Only the consequence: retrieval matches meaning, not exact strings. A passage is retrievable for a question when it covers the same concepts and names the same entities, which is why answer-first passages that restate the question's terms in their own first sentence do well without keyword repetition.