https://vapi.ai/ logo
Join Discord
Powered by
Fetch email during the call
# support
c
Hi everyone, I have created an assistant to catch the correct email address during the call. However after tying different models and modifying prompts, my assistant is still facing issues to get the correct email from the users. My email id has 3 i's in the end like xyziii@gmail.com. I have modified my prompt several time and also have asked confirmation of email characrter by character. However the assistant catches the incorrect email most of the times. It is important for me to catch the correct email address because I need to send the meeting link to the email id. The call id is => 019c1e1a-dba5-7bb8-ad84-685932dc3c70 Let me know hoe can I resolve this issue
c
Hi there, Thank you for your message. Our team is currently out of the office. We operate Monday through Friday, from 9:00 AM to 8:00 PM Pacific Standard Time (PST). We’ll get back to you as soon as possible during our normal business hours. If your message is urgent, please mark it accordingly or include “URGENT” in the subject line, and we’ll do our best to respond promptly. Warm regards, Vapi Customer Support Team
j
Hi, this is actually a very common issue with voice assistants, especially with emails that have repeated letters like “iii.” Most speech models tend to normalize or guess in those cases, which leads to wrong captures. A better approach is to use a guided spelling flow with smart confirmation and real-time validation instead of relying only on raw transcription. I can help you set this up properly so your assistant captures emails accurately every time. Are you currently doing any validation or correction after the email is captured? @Conquering-Digital
c
Thanks, Yes After the user has completed saying the email address the assistant repeates the email address character by character and ask for confirmation from the user. One more thing, mostly it the repeated characters that cause issue but sometimes it is the spelling of the name also that is incorrect. If the assistant can catch the email addressin in first or second attempt then it's fine otherwise if the iteration is more than 2-3 times then the email address becomes more and more incorrect
j
Thanks for explaining this, it really helps clarify what’s going on. What’s happening here is that after a few failed attempts, the system starts building on earlier mistakes, so each new try makes the email more distorted instead of fixing it. Just repeating it character by character isn’t enough you need a more controlled spelling flow, better confirmation logic, and a reset when errors happen. I’ve worked on similar setups and helped stabilize this so emails are captured correctly within one or two tries. I’d be happy to help you improve this, let’s discuss it more privately so I can look at your setup properly. @Conquering-Digital
c
Hi James, Currently I am the below prompt to fetch the correct email address of the user. Please have a look and share me your insights so that I can modify it accordingly.
## STEP 2 — EMAIL COLLECTION AND MAXIMUM-ACCURACY CONFIRMATION (CRITICAL) Ask: “Please tell me your email address character by character” ### Email Conversion Rules (STRICT) - Convert spoken words to characters exactly: Numbers: zero → 0, one → 1, two → 2, three → 3, four → 4, five → 5, six → 6, seven → 7, eight → 8, nine → 9 Symbols: at / at the rate → @ dot / point → . underscore → _ dash / hyphen / minus → - plus → + - Ignore “capital” or “small” indicators - Remove all spaces automatically - Never guess characters - Never treat numbers or symbols as letters - Listen carefully unless the user has spoken the full email address. - Do not interrupt the user if user is still speaking ### Domain Validation (Limited) - Correct only obvious mistakes: - gamil.com → gmail.com - hotmial.com → hotmail.com - If the domain is unfamiliar, ask the user to confirm it ### Structural Validation (MANDATORY) DO NOT PROCEED unless all checks pass: - Exactly one @ - At least one dot after @ - No dot immediately before or after @ - No consecutive dots - No spaces Save the validated email as: user_email --- ## STEP 3 — EMAIL CHARACTER-BY-CHARACTER CONFIRMATION (LOCK STEP) Say exactly: “Please verify your email if it's correct.” Then spell the email character by character, with short pauses: - Letters spoken individually - Numbers spoken as digits - Symbols spoken explicitly as: - “at” - “dot” - “underscore” - “dash” - “plus” Example: “r a h u l underscore s a i n i at g m a i l dot c o m” Then ask exactly: “Please confirm: should I send the meeting link to this exact email address?” - Wait for a clear yes or no - If the user says no: - Ask them to repeat the full email - Restart STEP 2 and STEP 3 - If yes: - LOCK the email - Do NOT modify it later and proceed to next step
j
Thanks for sharing this, your flow is actually well thought out and shows you’ve put real effort into improving accuracy. The main challenge is that it still depends a lot on the speech model getting every character right, and that’s where repeated letters and similar sounds usually break things. Once one small mistake happens, it tends to carry over into the next round, which is why the email gets worse after a few tries. A better approach is to add smarter fallback logic, confidence checks, and a proper reset instead of repeating the same loop. I’ve helped fix this in similar systems so emails are captured correctly within one or two attempts. If you’d like, send me a private message and we can go deeper into the best approach for your setup. @Conquering-Digital
s
getting email out of phone call is 10-20% accurate, we found a solution for that to get the correct email always
happy to help, let me know
t
@James Christopher I've sent you a message too, if that is ok
j
Sure proceed on sending me a private message @tommo154
c
Hi, This is a very common limitation with speech‑to‑text systems, especially when users have repeated letters in emails (like “iii”). There isn’t a setting that guarantees perfect capture here, so the solution is mostly about how you collect and confirm the email. What works best in practice: - Break the email into parts Ask for the part before 
@
, then confirm it, then ask for the domain. - Explicitly handle repeated letters When you hear something like “iii”, have the assistant ask: “Just to confirm, is that three separate letter i’s?” - Always read the full email back and confirm Do not proceed until the user explicitly confirms it’s correct. - Prompt users to spell the problematic part For example: “Please spell the last part letter by letter.” This is primarily a prompt and flow design issue, not a bug or model problem. Even the best transcribers can mishear repeated characters when spoken naturally.
PreviousNext
Powered by