But if it's just for the email and consistently, then I'd try prompt engineering the node or part of the assistant where it's reading out emails. Somehow vapi has terrible documentation on prompting but I've had really good luck metaprompting with chatgpt/gemini