Scenario
We're building an AI assistant using Copilot Studio. We filter our API data using a Custom Prompt Tool, then pass the filtered results into a Generative Answer node's Custom Data source. The Generative Answer node has prompt instructions to format a response using this custom data.
Result
Debug logs confirm the correct, filtered data is present in the Custom Data variable at the time of processing. The Generative Answer node visibly begins generating a correct response using that data, but once processing finishes, the response disappears and is replaced with "no information was found." The conversation then falls back to default orchestration, which returns a different, incorrect answer instead of using our data.
Environment
Copilot Studio agent, Generative Answer node with Custom Data source populated from a Custom Prompt Tool's structured output.
Troubleshooting efforts
Verified via debug logging that the row count and content of the Custom Data variable are correct immediately before the Generative Answer node runs. The issue occurs consistently, not intermittently.
Question
Why does the Generative Answer node discard a correct, already-generated response and replace it with a "no data found" fallback, even though the custom data is confirmed correct at that point? Is there a way to stop this fallback and just use the generated result?
Thank you!