Unable to label generated fields in Content Understanding Studio

Question

Unable to label generated fields in Content Understanding Studio

Joe Plumb 0

I have a Content Understanding Studio project where I need to tune the models further and provide labelled examples to the service to improve accuracy. I am able to label fields that use the 'Extract' method, but not those set to 'Generate'. Is this expected behaviour?

SRILAKSHMI C 18,040 Reputation points Microsoft External Staff Moderator

2026-04-16T14:57:56.46+00:00
Hello Joe Plumb,

Thank you for your question.

This is expected behavior in Content Understanding Studio.

In Content Understanding Studio, labeling capabilities depend on how the field is configured:

Extract fields These fields are based on direct extraction from the source document or transcript (e.g., specific text spans or key-value pairs). Because there is a clear source location, the Studio provides a manual labeling experience, allowing you to select and tag the relevant content.

Generate fields These fields are produced using generative AI (LLM-based inference) and are not tied to a fixed span in the source content. Since there is no explicit region to associate with a label, the Studio does not provide a labeling option for these fields.

The built-in labeling UI only supports “Extract” fields

“Generate” fields are not labelable within the Studio by design

This is because their outputs are synthesized rather than directly extracted

Recommended approach for “Generate” fields

If you would like to improve the accuracy of generated fields, the recommended approach is:

Collect ground-truth examples externally (input - expected output pairs)

Use these for:

Prompt tuning / prompt refinement

Model fine-tuning workflows (if applicable)

Additionally, you may consider:

Refining the instructions or schema definition for the generated field

Switching to Extract-based fields where deterministic labeling and training is required

Here are some docs that cover the current labeling support and setup:

“I can’t label my data” – troubleshooting common labeling issues (CORS, storage config) https://docs.microsoft.com/azure/applied-ai-services/form-recognizer/quickstarts/try-v3-form-recognizer-studio#additional-prerequisites-for-custom-projects

Setup and configuration of labeling project – prerequisites for custom projects in the Studio https://docs.microsoft.com/azure/applied-ai-services/form-recognizer/quickstarts/try-v3-form-recognizer-studio#configure-cors

Local labeling (sample labeling tool) – for on-prem or VNet-isolated scenarios https://docs.microsoft.com/azure/applied-ai-services/form-recognizer/deploy-label-tool

Azure Content Understanding overview – explains Extract vs. Generate methods https://learn.microsoft.com/azure/ai-services/content-understanding/overview?wt.mc_id=knowledgesearch_inproduct_azure-cxp-community-insider#key-components-of-content-understanding

Content Understanding glossary – definitions for generation methods https://learn.microsoft.com/azure/ai-services/content-understanding/glossary?wt.mc_id=knowledgesearch_inproduct_azure-cxp-community-insider

Best practices: generating labeled datasets – tips on building your ground-truth for training https://learn.microsoft.com/azure/ai-services/document-intelligence/train/custom-labels?view=doc-intel-4.0.0&wt.mc_id=knowledgesearch_inproduct_azure-cxp-community-insider#start-by-identifying-the-fields

I Hope this helps. Do let me know if you have any further queries.

Thank you!
SRILAKSHMI C 18,040 Reputation points Microsoft External Staff Moderator

2026-04-17T08:23:38.4066667+00:00

Hi @Joe Plumb

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!