The voice sounds nice, but successful runs aren't that interesting. If it gets a question it doesn't understand, what does it do, and how does it report it back

The voice sounds nice, but successful runs aren't that interesting. If it gets a question it doesn't understand, what does it do, and how does it report it back to the user? It seems like the user is likely to get a certain number of confused recording sent back to them when it fails, and then get stuck manually calling back and explaining what happened.