Hello ,
Thanks for the response. I am already working in the web app version of copilot studio. I've looked through the activity logs at the conversations and the debug box does not show in the conversation transcript at all. Also it honestly seems like it has resolved itself with yesterday despite me not making any major changes. I'll answer everything to the best that I can so others can hopefully find it helpful.
Yes I am using generative answer nodes, but this has also occurred when no topic has been created and the agent is responding using generative orchestration. The structure of the topics I am using is very simple; I only really created them to utilize slot filling for some things.
Agent chooses trigger -> Generative Node. It has an input for the specific intent of the query and an output for the response.
Due to my org going through a transition of how we are storing our work instructions and other data, all knowledge sources are currently just PDFs that have been directly uploaded to the agent. I had been experimenting with only allowing specific knowledge sources but this did not seem to have an effect. I tried setting moderation level to the lowest setting as well, but this again did not have an effect. The only other thing I can think of was some very minor adjustments to knowledge source descriptions, but since the I can see the agent in the test pane generating a perfectly accurate response prior to it filtering the message and claiming unknown intent I feel like something else is happening.
I wish I had a better explanation but part of me just thinks that the model was acting up. The problems started propping up right when MS released GPT 5, depreciated the older versions and then subsequently walked back the decision after the backlash. With how confusing MS docs can be at times though I'm not even sure I understand that situation correctly, or if Copilot Studio was even affected by the situation at all.
Thanks for your time,
Jason
