Amazon Web Services: Retrieval Instructions Control How Answers Are Written, Metadata Filters Control What Is Retrieved—A Multi-Tenant Customization Approach for Shared Knowledge Bases
The Amazon Web Services Machine Learning Blog introduces the retrieval instruction feature of Amazon Bedrock Managed Knowledge Base. The core distinction is that instructions set each tenant's voice, format, retrieval guidance, and answer policy, while metadata filters control which documents each tenant can retrieve; the two operate independently. The article also notes that answer policies are best-effort—if restricted documents are retrieved, answers may still reference their mechanisms or record IDs, so data isolation must rely on filters rather than policy text.
The same product documentation and policy files often serve different audiences: customers in three sales regions need their own languages and date formats, while internal support teams need short ticket notes. Previously, implementing this differentiation on Amazon Bedrock Managed Knowledge Base required developers to disable response generation, call models themselves, and maintain prompts in every application. An article published by the Amazon Web Services Machine Learning Blog in November 2026 introduced a new mechanism: retrieval instructions—a versioned managed resource referenced by each request, eliminating the need to duplicate maintenance across applications.
The article's core distinction is noteworthy: retrieval instructions change how answers are written but do not limit what is retrieved. Instructions have four fields—persona sets identity, tone, and language; answerFormat specifies structure, date and currency formats, and citation style; retrievalGuidance directs the agent on how to interpret questions and prioritize document searches; answerPolicy determines whether to answer and how to handle out-of-scope questions. Among these, persona, answerFormat, and answerPolicy only take effect when generateResponse is true, while retrievalGuidance applies regardless of response generation. However, the author explicitly warns that these fields are not authorization or data-isolation boundaries. To ensure different tenants see only their own documents, metadata filters or separate knowledge bases must be used to control the retrieval scope.