Hi everyone, I’m troubleshooting an issue in Microsoft Copilot Studio.
Last week, my test agent’s Create generative answers step generated responses correctly from a prompt where I inject a variable containing a JSON table.
Starting this week, the same step returns an empty generative response.
When I test the agent in chat, I briefly see a message like “Edit Knowledge Base” flash for a second, then disappear. After that, the generative answer comes back blank - as if it tries to use Knowledge first, even though I’m not using a knowledge base.
Environment
-
Corporate tenant (possible governance/policy changes)
-
This worked previously in the same agent/flow; stopped around Monday (this week)
What I’m trying to do
Generate an answer purely from the prompt + injected JSON data (no KB sources).
What I tried
-
Re-checked the node configuration and prompt text
-
Retested multiple times in the test chat (Preview)
-
(If applicable) toggled/removed knowledge sources / verified none are intentionally used
Questions
-
Does the “Edit Knowledge Base” flash indicate the generative answers node is being forced to check Knowledge sources first?
-
What settings/policies most commonly cause empty outputs in this node (tenant restrictions, DLP/Power Platform policies, permissions, AI features disabled, etc.)?
-
What should I ask internal support/admins to verify (specific admin center / Power Platform / Copilot Studio settings)?

Report
All responses (
Answers (