AI-Generated Descriptions Disappear in Edit Mode: Root Cause and Workaround – DataHub Customer Support Portal

Area: Product Issues

Sub-Area: AI Documentation / Description Editing

Issue

When a DataHub asset has an AI-generated description visible in the Summary/Documentation panel, clicking the edit (pencil) icon opens the description editor with a completely blank text field. The AI-generated text that is clearly visible in the read-only panel does not pre-populate the editor, forcing users to manually copy and paste content before they can refine or extend it. This behavior affects every asset whose documentation was produced solely by AI automation and has never been manually authored or explicitly published through the edit dialog. At scale, this makes AI-assisted curation workflows impractical for large asset inventories.

Error Messages

No error message is displayed — the editor silently opens with an empty string.

You Might Be Asking

Why does the Documentation edit box appear empty when there is clearly text showing in the panel?
Why does re-generating with AI inside the editor and publishing fix the problem for future edits?
Is there a way to bulk-fix assets that are stuck in this state without re-generating every description?
Which DataHub version contains the fix for this issue?

Solution

Understand the root cause.
AI-generated descriptions and manually authored descriptions are stored in separate fields internally. The utility function getAssetDescriptionDetails() (located at datahub-web-react/src/app/entityV2/shared/tabs/Documentation/utils.ts) filters out documentation entries flagged as inferred=true unless the caller explicitly passes an enableInferredDescriptions flag. The read-only Summary panel correctly passes this flag and therefore displays AI content. Prior to the fix, the hook that supplies the initial content to the Edit Description modal (useDescriptionUtils.ts) omitted the flag, causing the editor to receive an empty list and render a blank field. This affects any asset where:
- The editable description aspect is NULL
- The ingested schema description is NULL
- The only documentation present is flagged as inferred (AI-generated)
Apply the immediate workaround (per-asset).
Inside the Edit Description dialog, click Generate with AI to produce a new AI description, then click Publish. This saves the text as a user-authored entry, which the editor can correctly pre-populate on all subsequent edits. Note that re-generation may produce text that differs slightly from the original AI output.

Apply a bulk workaround via the API (optional, for large inventories).

If you have a large number of affected assets, you can programmatically promote AI-generated descriptions to editable descriptions using the DataHub REST API. This preserves the existing AI-generated text exactly and makes the edit boxes fill correctly without re-generation. Use the metadata write endpoint with appropriate back-off to avoid performance impact:

# Example: promote an AI-generated description to an editable description
# Replace placeholders with your environment values

curl -X POST \
  "https://<your-instance>.acryl.io/api/gms/aspects?action=ingestProposal" \
  -H "Authorization: Bearer <your-api-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "proposal": {
      "entityType": "dataset",
      "entityUrn": "urn:li:dataset:(urn:li:dataPlatform:<platform>,<dataset-name>,<env>)",
      "aspectName": "editableDatasetProperties",
      "changeType": "UPSERT",
      "aspect": {
        "contentType": "application/json",
        "value": "{\"description\": \"<paste-ai-generated-text-here>\"}"
      }
    }
  }'

Add rate-limiting or sleep intervals between calls when processing large batches to avoid overloading the metadata service.

Upgrade to v2.2.x or later for a permanent fix.
Engineering merged a fix in PR #11678 that corrects useDescriptionUtils.ts to pass the enableInferredDescriptions flag and adds a contentKey to re-mount the editor once description data arrives (covering a secondary race condition on the ?editingDescription=true entry path). After upgrading, the edit dialog will correctly pre-populate from AI-generated content when no manually authored description exists. Assets that had empty editable descriptions saved on top of AI content may still show blank — in that case, remove the empty editable description or paste the correct text in manually.

Additional Notes

This bug was confirmed present in DataHub Cloud versions up to and including v2.1.4-cloud. The fix is included starting with v2.2.x. The fix does not retroactively correct assets where a user previously published an empty description on top of the AI-generated text — those will need to be identified and corrected manually or via API. The read-only Documentation panel is not affected and always displays AI-generated content correctly regardless of version. The workaround of re-generating inside the editor changes the stored text, so if exact preservation of the original AI output is important to your organization, prefer the API-based promotion approach or wait for the platform fix before editing at scale.

Issue

Error Messages

You Might Be Asking

Solution

Additional Notes

Related Documentation