Share via

Content Comprehension: The model stopped returning values and now only outputs data types

Sergio 0 Reputation points
2025-12-05T10:21:54.7866667+00:00

Hi,

I’m using Azure AI Services – Content Comprehension with a custom analyzer to extract structured data from invoice PDFs.

Until recently, the API was correctly returning the values for each field (for example: CompanyName, InvoiceNumber, TotalAmount, etc.). I did not change my schema or my request code, but now the response has changed and it only returns the data types, not the actual values.

Here is an example of the kind of response I’m getting now (sample data is synthetic and redacted):

{
  "status": "Succeeded",
  "result": {
    "contents": [
      {
        "path": "invoice",
        "fields": {
          "CompanyName": {
            "type": "string"
          },
          "InvoiceNumber": {
            "type": "string"
          },
          "SaleDate": {
            "type": "date"
          },
          "TotalAmount": {
            "type": "number"
          },
          "Items": {
            "type": "array"
          }
        },
        "kind": "document",
        "mimeType": "application/pdf"
      }
    ]
  }
}

Previously, I was getting values like this:

"CompanyName": {
  "type": "string",
  "value": "ACME SUPERMARKETS S.A."
},
"InvoiceNumber": {
  "type": "string",
  "value": "F-2024-00123"
}

But now I only get:

"CompanyName": { "type": "string" }

No value or content is being returned anymore.

Azure AI Document Intelligence
0 comments

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. Anshika Varshney 4,965 Reputation points Microsoft External Staff Moderator
    2025-12-17T14:45:22.2366667+00:00

    Hi Sergio,
    Thank you for reaching out on the Microsoft Q&A.
    Why Your Content Comprehension Suddenly Stopped Returning Values (Only Types)

    You’re seeing only field types (e.g., "type": "string") and no "value" fields even though the request succeeds and without any changes to your code or schema because the backend pipeline used by Content Comprehension has changed recently. The service has moved from a preview model stack to the new GA (General Availability) pipeline that requires model defaults to be configured on your AI resource. Microsoft Learn

    In the older (preview) path, analyzers could run and extract full values without requiring any extra model configuration. But in the newer GA backend, the engine depends on your Content Understanding resource having model deployments attached (e.g., GPT and embedding models) to actually extract and populate field values.

    If those defaults are not set, the analyzer still parses the document successfully which is why status is "Succeeded" but it only builds the field schema and never runs the extraction step that produces values.

    How to Fix It

    Here’s what you need to do so your analyzer returns full values like before:

    1.Use the GA API Version

    Make sure your calls specify the GA API version:

    api-version=2025-11-01

    If you’re still invoking the preview version (2025-05-01-preview), the preview backend doesn’t work with GA analyzers and will only return types.

    2.Attach Required Model Defaults on Your Content Understanding Resource

    In the Azure portal:

    Go to your Content Understanding / Foundry AI resource

    Open Settings → Resource Defaults

    Attach at least the following models:

    gpt-4.1

      **gpt-4.1-mini**
  
     **text-embedding-3-large**
    1. Enable “Auto-deploy” so that these models are auto-deployed if not already.

    Without these model deployments, the GA backend will parse documents into types but won’t extract any field values which matches exactly what you’re observing.

    3.Don’t Use the Old Foundry “Test” Panel

    The classic “Test” tab in the Foundry UI still hits the preview API. Even if it appears to work, it doesn’t use the GA pipeline and won’t return values with GA analyzers. Instead, test using Content Understanding Studio or direct API calls using the GA version (2025-11-01).

    4.Recreate or Re-Train the Analyzer Under GA

    Even if your schema hasn’t changed, preview analyzers don’t automatically migrate to the GA pipeline. So:

    • Clone or recreate the analyzer

    Publish it using the GA API

    Then test it again

    This ensures the analyzer runs with the correct backend that uses your attached models.

    I hope this has been helpful! Please let me know if there are any remaining questions or additional details, I can help with, I’ll be glad to provide further clarification or guidance.

    Thankyou!

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.