Copilot Studio Not Retrieving Metadata Fields from Azure AI Search Knowledge Base

Question

Copilot Studio Not Retrieving Metadata Fields from Azure AI Search Knowledge Base

Rodríguez, Alberto 25

We have integrated Copilot Studio with Azure AI Search as our knowledge base. The search functionality is working correctly and returns relevant content; however, metadata fields (such as document title, identifier, and other custom fields) are not being included in the responses.

These metadata fields are properly configured and marked as retrievable in Azure AI Search, but Copilot Studio does not seem to return or utilize them when generating answers.

Is there a way to configure Copilot Studio to include and leverage these metadata fields in its responses so that we can generate more detailed and context-rich answers?

0 comments

Answer accepted by question author

0 additional answers

Your answer

Answer 1

Ashlesha Pawar (Tata Consultancy Services Limited) 390 Microsoft External Staff

Rodríguez, Alberto

This behavior appears to be a current limitation of Microsoft Copilot Studio when using Azure AI Search as a native knowledge source.

Although Azure AI Search fields can be configured as retrievable, Copilot Studio’s built-in knowledge experience primarily consumes document/snippet content and limited citation information, rather than exposing arbitrary metadata fields (for example: custom IDs, tags, categories, or other index attributes) directly to generated answers.

Microsoft documentation currently does not describe any supported configuration for mapping Azure AI Search metadata fields into Copilot Studio response generation. The documentation instead points to using custom knowledge sources for advanced retrieval and orchestration scenarios.

Rodríguez, Alberto 25 Reputation points

2026-05-07T07:06:31.2066667+00:00

With the Semantic Hybrid Search tool I’m able to retrieve the metadata correctly, but when I pass it to the generative answers option, the context is not being included properly in the response. As a result, the generated answer does not use that information as expected.

Is there any alternative way to instruct the generative answers tool to use this knowledge/context more effectively?
Ashlesha Pawar (Tata Consultancy Services Limited) 390 Reputation points Microsoft External Staff

2026-05-08T12:22:14.19+00:00
Rodríguez, Alberto,

What you’re seeing is a current limitation of Copilot Studio’s Generative Answers. Even though Azure AI Search returns the metadata correctly, Generative Answers mainly uses the document content/chunks and may ignore custom metadata fields during grounding.

The most reliable workaround is to include important metadata directly inside the indexed content itself, for example:

Title: Invoice Policy Document ID: DOC-1045 Category: Finance Employees must submit invoices within 30 days...

Another option is to avoid relying only on Generative Answers and instead:

retrieve results with Semantic Hybrid Search,

build a custom prompt including metadata + content,

then pass that to Azure OpenAI/LLM.

This gives you full control over how metadata is used in the final response.