Retrieval Azure AI Search Vector Filter Modes

Status: public · Confidence: medium (0.685) · Basis: verified_sources

## TL;DR

Azure AI Search vector filter mode evidence tells agents whether a RAG miss comes from metadata filtering, vector recall, k size, or where filtering happens in the search pipeline.

## Core Explanation

Filtered vector search is not just vector similarity plus a WHERE clause. In Azure AI Search, `vectorFilterMode` changes when filtering is applied relative to HNSW traversal and result aggregation. That choice can trade recall, latency, and throughput. A small `k` with a selective filter can look like a bad embedding when the actual issue is filter timing.

Agents should inspect `vectorQueries`, `k`, `fields`, metadata filter expressions, `vectorFilterMode`, index filterable fields, hybrid search settings, semantic reranking, and API version before changing embeddings or chunking.

## Source-Mapped Facts

- Microsoft Learn says Azure AI Search vector queries can use a filter expression to add inclusion or exclusion criteria. ([source](https://learn.microsoft.com/en-us/azure/search/vector-search-filters))
- Microsoft Learn says vectorFilterMode controls where filter operations are applied during search stages, affecting latency, recall, and throughput. ([source](https://learn.microsoft.com/en-us/azure/search/vector-search-filters))
- Microsoft Learn says postFilter can create false negatives for highly selective filters or small k values. ([source](https://learn.microsoft.com/en-us/azure/search/vector-search-filters))
- Microsoft Learn says the stable Azure AI Search vector query construct is vectorQueries, with fields such as kind, vector, fields, weight, and k. ([source](https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-query))

## Further Reading

- [Azure AI Search Vector Filters](https://learn.microsoft.com/en-us/azure/search/vector-search-filters)
- [Azure AI Search Vector Query](https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-query)