You store embeddings in a property named embedding in an Azure Cosmos DB for NoSQL...

The query should limit the number of returned results by using TOP 5 together with ORDER BY VectorDistance(...) . Azure Cosmos DB for NoSQL provides the VectorDistance() system function for comparing a stored vector with an input query vector. Microsoft documents that a vector similarity query should order the results by VectorDistance so that the nearest matches are returned first.

A representative query is:

SELECT TOP 5

c.id,

VectorDistance(c.embedding, @embedding) AS SimilarityScore

FROM c

ORDER BY VectorDistance(c.embedding, @embedding)

Microsoft specifically recommends using a TOP N clause for vector search queries because returning more results than necessary increases RU consumption and latency. In this scenario, TOP 5 directly satisfies the requirement to return only the five most similar documents.

GROUP BY is used for aggregation and does not rank vectors by similarity. DISTINCT removes duplicate projected values but does not restrict results to the five nearest vectors. Offset pagination is unnecessary and can introduce additional query cost when only the highest-ranked five results are required.

Therefore, the correct action is to limit the number of results .

Study Guide references: Azure Cosmos DB for NoSQL → VectorDistance() ; vector search; ORDER BY ; TOP N ; RU and latency optimization.