What is an embedding?
An embedding represents information, such as text or an image, as a numerical vector so a system can calculate similarity between items.
In semantic search, the system can compare the query's vector with document vectors to find material related in meaning, even when the wording differs.
Why different wording can lead to a relevant result
A query about “reducing AI inference costs” might need articles about quantization or inference chips. Those pages may be useful even if they do not repeat the exact query.
Cohere's documentation describes embeddings for semantic search, clustering and classification. A search service may use such representations as part of retrieval; the service's actual implementation still matters.
Similarity is a relevance signal, not a fact check
Nearby vectors do not establish that a document is trustworthy, current or correct. Two passages can discuss the same subject while disagreeing about a fact.
In a RAG workflow, examine the retrieved candidates and their applicability before relying on the generated answer. Keep relevance, source quality and faithfulness to the source as separate checks.
Sources and scope
Source-language entry records a source check on 2026-10-21. This adaptation preserves that provenance and does not represent a new audit of the cited pages.