Trust markers
Help users understand what is happening below the surface and trust the results returned or the underlying model
Caveats
Inform users about shortcomings or risk in the model or the technology overall
Controls
Manage the flow of information or pause a request mid-stream to adjust the prompt
Feedback
Signal expectation gaps or errors in the model – but is that clear to the user?
Footprints
Give users the ability to trace the relationships between sources and results