Azure AI Foundry, Content Understanding returning "Not Found"

Hello Garin Evans,

Thanks for raising this question. The "Not Found" responses from Azure AI Foundry's Content Understanding service, where it detects the correct number of products (9 in your case) but returns empty fields in the raw JSON output, typically point to a schema mismatch, indexing lag after recent changes, or a transient issue in the preview API—especially since you mentioned no schema updates and it was working until last week. This is a common preview quirk where the analyzer (auto-labeling-model-1762962853246-223) parses the document structure but fails to map values to fields like Title/SubTitle/Description/ImageURL, often due to model retraining delays or content parsing thresholds. Let's troubleshoot and resolve step-by-step.

Step 1: Verify Schema and Analyzer Configuration

• Go to your AI Foundry project in the Azure portal (West US region) > Content Understanding > Select your analyzer (auto-labeling-model-1762962853246-223).
• Check the schema: Under Schema tab, ensure fields (Products array with Title, SubTitle, etc.) match your HTML structure exactly—e.g., Title should target <h1> or .product-title selectors if using custom prompts. If fields are defined as "string" without extraction rules, add explicit prompts like "Extract product title from <h2> tags."
• Reindex: Click Rebuild index or Train model to force retraining—wait 5-15 minutes, then test with a sample document. Preview API changes can cause silent desyncs.

Step 2: Test Document Input and API Calls

• Validate input: In the project > Test tab, upload a fresh HTML file and run analysis. If raw response shows empty valueObject, the issue is in parsing—ensure Markdown conversion preserves HTML tags (use kind: html in API if supported).
• Check API parameters: If calling via REST/SDK, confirm payload includes the correct analyzerId and api-version=2025-05-01-preview. Example curl test:

  
    curl -X POST "https://your-foundry-instance.cognitiveservices.azure.com/contentunderstanding/analyze?api-version=2025-05-01-preview" \
  
    -H "Ocp-Apim-Subscription-Key: your_key" \
  
    -H "Content-Type: application/json" \
  
    -d '{
  
      "analyzerId": "auto-labeling-model-1762962853246-223",
  
      "documents": [{"content": "your_html_here", "kind": "html"}]
  
    }'
  
  

  • Look for warnings in the response (your JSON shows empty array)—if present, they indicate parsing failures (e.g., "No matching elements found").
• Poll status: For async runs, ensure you're retrieving results after "Succeeded" (your output shows status: Succeeded, so sync calls are fine).

Step 3: Regional and Service Health Check

• Review service health: In Azure portal > Service Health > Filter for AI services/Content Understanding > Check for West US incidents last week (e.g., model deployment delays). If any, wait for resolution or switch to East US 2 temporarily.
• Quota/Throttling: Under your resource > Quotas > Ensure no exceeded limits for calls (default 1,000/day in preview). Monitor metrics for "Failed Requests" spikes.
• Recreate Analyzer: As a workaround, duplicate the analyzer (Project > Content Understanding > Clone) with the same schema, assign a new ID, and test—deletes/rebuilds often clear stale states in preview.

Step 4: Advanced Fixes and Support

• Custom Prompts: Edit analyzer settings > Add extraction prompts like "Identify products in HTML, extract Title from <div class='product-name'>, ImageURL from <img src>." Retrain and retest.
• SDK Fallback: If using SDK, update to latest (@azure/ai-content-understanding 1.0+ preview) and log full responses—sometimes raw JSON reveals hidden errors.
• Escalate: If persists, open support ticket: Portal > Help + support > New request > Technical > AI services > Content Understanding > "Not Found with empty fields despite Succeeded status" > Attach raw JSON, analyzerId, and sample input. Response: 1-2 days; reference preview limitations for priority.

This should get your product scraping back on track start with schema rebuild, as it's the most common culprit. Kindly share your feedback trying the above steps. If it works accept this answer.

Best Regards,

Jerald Felix

Hi, thanks for the suggestions.

I can confirm the schema has not changed since the data was being returned successfully, to follow your advice I have added specificity to the field definitions and still it returns nothing.

I have retrained and rebuilt the analyzer, still nothing.

I have uploaded fresh html files. Nothing.

I have checked issues in the region where my analyzer is built, there are none apparently, though Content Understanding does not appear as an option to select. I wonder whether it's not reported given it's in preview?

I am using the Test options within the Foundry portal, and so nothing required to verify the SDK or the API.

I wonder whether something has fundamentally changed with the model Content Understanding is using?

Hi Garin Evans

When GA released (API version 2025‑11‑01), Microsoft changed the pipeline so that Content Understanding no longer runs its own internal model stack.

It now requires you to attach Foundry model deployments as Resource‑Level Defaults. If these defaults are not set, CU will parse your document but return empty fields with no values, exactly what you're seeing.

The Foundry “Test” panel in the portal still uses the Preview backend

The current Foundry UI has not fully switched to the GA pipeline, so:

• You may be testing your analyzer using the old preview API (2025‑05‑01‑preview),
• That preview version does not work with the new GA analyzers,

Result → you see zero extracted values, even though schema is correct.

HTML extraction on the Preview path has known issues

• When preview is accidentally used:
• HTML selectors stop applying
• Tables and multi-node arrays return only empty shells
• Even previously working schemas return nothing

This matches your exact symptoms.

The reason you don’t see “Content Understanding” as an option

GA moved the control plane behind:

• Content Understanding Studio (new UX)
• Foundry resource defaults

The classic Foundry UI hasn’t been updated yet, so things look broken even though GA is live.

This is expected right now during the transition.

How to fix it

Check which API version you are actually hitting

If your Test panel shows 2025‑05‑01‑preview, you are on the wrong version.

Correct version must be:

api-version=2025-11-01 

Content Understanding → Settings → Add Resource → Attach your Foundry Resource

Then set:

Check “Enable autodeployment” if models aren’t already deployed.

Without these three defaults, GA returns empty results for every analyzer.

Test using Content Understanding Studio (NOT the Foundry Test tab)

Because the Foundry portal is still pointing to the preview backend, your GA analyzer returns nothing there.

Studio uses the correct GA backend.

Recreate the analyzer on GA, or copy from preview

Preview analyzers don’t fully migrate even if schema is identical.

Create a new one under GA and test again.

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

Thank you!

Hi Garin Evans

Did you get any chance to review the above response. Thank you!