On a recent call, the agent addressed the user by a made up name despite the user never providing their name, causing a lot of frustration. I have nothing in my prompts telling the agent to address the user by name, so I don't know why the agent would do that. How do I prevent against this in the future?

Hi @pymlu16

call_b2115a866edbc6f7295dbf7405e

Which model are you using

GPT5.2

You should use 4.1

why?

Try using 4.1 or 4.1 mini as they are more optimized

okay thanks. And while I have you, is there a better transcription model we can use that's not Deepgram or Azure? they are very bad at transcribing the user's name

Azure is better and for more better transcription you can use boosted words , or you can also try switching the transcription mode to optimize for accuracy