Hi,
This is expected behavior when you use data sources in the Global “Generative AI” settings in Copilot Studio.
What’s happening:
When global data sources are enabled, the orchestrator takes over and directly handles the generative answer. In this flow, the system topic (like “Conversational Boosting”) is bypassed, which is why your Adaptive Card and custom logic are not executed.
This is by design — the orchestrator prioritizes answering the query and ends the turn immediately after generating the response.
Why your approach isn’t working:
Even if you modify the “Conversational Boosting” topic or move data sources locally, the orchestrator can still override and skip topic execution when it determines it can answer directly.
How to handle this:
1. Capture user input before orchestration
You need to store the user query early in the conversation, before the generative answer is triggered.
For example, create a variable at the start of the conversation and store:
System.Activity.Text
2. Use a fallback or post-processing topic
Since you cannot force execution after the generative answer when using global data sources, you can:
- Trigger a follow-up topic after the response
- Or guide the conversation to a feedback step explicitly (e.g., “Was this helpful?”)
3. Avoid relying on Conversational Boosting topic
This topic is not guaranteed to execute when orchestration is active. Instead, design your flow using:
- Custom topics
- Explicit prompts for feedback
4. Alternative approach (recommended)
If capturing feedback immediately after every answer is critical:
- Disable global data sources
- Handle generative answers inside a topic using the “Create generative answers” node
This ensures full control over the flow, including showing Adaptive Cards right after the response.
Key takeaway:
There is currently no supported way to force execution of additional logic (like Adaptive Cards) immediately after a generative answer when using global data sources, because the orchestrator bypasses topic-level control.
Suggested approach:
Either capture the user input before orchestration starts, or move to a fully controlled topic-based generative setup where you can guarantee execution of your feedback flow.
Hope this helps,
If this helps resolve your issue, please consider marking the response as Verified so it can help others facing a similar scenario.
If you found this helpful, you can also click “Yes” on “Was this reply helpful?” or give it a Like.
