Embeddings

Embeddings are numerical representations of text that let AI systems measure semantic similarity between a query and a set of candidate passages, forming the underlying mechanism that makes semantic search and most retrieval-augmented generation possible in the first place.

Why it matters

Understanding embeddings clarifies why matching a query's exact keywords matters less to AI retrieval than it did to traditional search: a passage can be retrieved because it's semantically close to a query's meaning even without sharing its literal words, the mechanical basis for semantic search itself.

Related terms

• Semantic Search

  Semantic search is retrieval based on the meaning and intent behind a query, rather than exact keyword matching. It uses vector representations of text to find content that's conceptually related even when the actual wording is completely different.

• Retrieval-Augmented Generation (RAG)

  Retrieval-Augmented Generation (RAG) is the architecture behind most AI search: the system retrieves a set of relevant documents or passages at query time, then generates its answer grounded in that retrieved material, rather than purely from what it learned during training.

• Chunking

  Chunking is how AI systems split content into smaller passages, typically 200 to 500 tokens, before indexing and retrieving it, meaning a page is rarely evaluated as one continuous whole but as a set of independent segments competing separately for retrieval.