Something I've been dealing with that has got to be the most frustrating thing ever is I'll have the ai assistant ask for an email and the user will give it and then the AI will say "what was that?" I have endless amounts of directions in my prompts telling it exactly what to do and exactly what not to do. I have tools added to say things but for some reason there is some kind of fallback in vapi that is making it say "what was that?" None of my agents are even able to collect an email right now. I checked Retell to see if it did the same thing and it didn't even without the tools just the prompt. Is there some kind of fallback set in Vapi with getting emails? And if there is, is there something we can change to make it just say nothing or "" instead of "what was that?" ?

Have you tried creating a tool that collects user email?

Yes, that's what I've been doing. It fixes the problem maybe 1/5 of the time.

Can I have access to the vapi logins so I can find out where the issue is coming from

I have to many clients in my vapi account. However, I can give you the prompt and all my info if that works for you?

The problem is once I debug the issue I'll need to test the assistant as well and I can't do that without access

I will give you all the info from my vapi assistant

Alright then DM

A tool is not required for this type of interaction, though it can help sometimes. Really the best way to resolve this is by prompt engineering. Writing clear instructions for the assistant to collect information from the user can help automate this task: "Your task is to politely ask the caller for their email address, confirm that you understood it correctly, and save it. Ask for the email address in a natural and professional tone. Example: “Could you please provide the best email address where we can reach you?” When the caller responds, repeat the email address back clearly, spelling it out if necessary. Example: “I heard john.smith at gmail dot com. Is that correct?” If the caller says yes, confirm: “Great, I’ve saved your email address.” If the caller says no, politely ask them to repeat the email address. Example: “Thank you, could you repeat it once more slowly so I can make sure I get it right?” If the caller hesitates or doesn’t provide an email, gently nudge them once more. Example: “We’ll need an email address so we can follow up with you. Would you like to use your personal or work email?” Do not proceed to the next step until a valid email format has been collected and confirmed."

I've tried this as well @kyle. It always likes to just add a random "What was that?" in there so weirdly. Also, I don't think this is the best way to do it because normally within the call the assistant hears the email wrong. But then when I send the transcript through make.com it will pick it up correctly.

Any progress with this issue?

Another community member mentioned that using Azure OpenAI without a key prevents this issue, but I have yet to test this. Might be worth a try.

I am using Azure Open AI but I have a key. However, I was having problems early this month with my key inside of vapi not connecting properly. Maybe that has something todo with it.

message has been deleted

If you would like, we can remove the key from your account if it is currently connected. Just provide your org id and we will proceed with removal