Power Platform Community Forum Thread Details

I built a FAQ agent in Copilot Studio. The Create generative answers node returns "no relevant information was found" for every query, even though the flow runs correctly.

What I have already verified:

Input is correctly bound to my question variable (UserQuestion)

"Send a message" on the generative answers node is turned OFF (no duplicate output)

Orchestration is set to Generative, content moderation is Medium

Knowledge sources (a public website source and a SharePoint source) are in "Ready" status

I tested with an in-scope query that matches the website source's scope — still 0 results

I turned ON "Allow the AI to use its own general knowledge" — still returns "no information found"

Observed behavior: the ideal answer appears for a split second, then is replaced by the "no information found" message.

Topic checker shows errors: PowerFxError / "identifier ... not recognized in expression ''".

The most confusing part is that this is inconsistent — it works sometimes and fails other times. The exact same agent and the exact same question can return a proper answer in one attempt and "no information found" in another, with no configuration change in between. We have also seen it behave differently across environments — working in some and failing in others. Multiple engineers have reviewed it and none of us can pin down a reproducible trigger. This intermittency makes me suspect a tenant-level capacity/license limit or a backend/service-side factor rather than a topic configuration issue.

Has anyone seen this pattern where generative answers intermittently returns nothing — even with general knowledge enabled? Any pointers on what causes the inconsistency would be appreciated.

スクリーンショット 2026-06...

Your file is currently under scan for potential threats. Please wait while we review it for any viruses or malicious content.

Categories:

Topic creation & management

Based on what you described, it does not sound like a tenant capacity/license limit as the primary cause. Copilot Studio documents that when quotas are actually hit, users see an explicit failure notice, whereas your symptom is an answer flashing briefly and then being replaced by the node’s 'no information found' path.

If the node input, custom response handling, or downstream expression is intermittently evaluating with an invalid variable reference, the generative answer can appear momentarily and then get overwritten by fallback behavior. In other words, the flow may be reaching the node, but the final result handling is still failing in-topic.

A few other things are worth checking as well:

SharePoint auth/permissions: SharePoint knowledge sources run with the agent user’s Microsoft Entra ID and only return content that the current user can access, so inconsistent results can happen across users/environments if auth or permissions differ.

Source retrieval behavior: Public website sources rely on Bing-grounded retrieval, and SharePoint retrieval depends on search/indexing. “Ready” only means the source is available to test; it does not guarantee every query will return results. Microsoft also recommends isolating sources during testing and using site:<url> checks for website coverage.

Content moderation / follow-up blocking: Generative answers can return nothing if moderation blocks the response, and with generative orchestration the “Allow ungrounded responses” setting only helps when the model answers without using a source/tool in that turn.