Hello, so we have a custom GPT that we setup to authenticate with our Thinkific course platform. We used Vercel to do this. Everything is working great, authentication works, well it works when the GPT actually asks for the email to verify.
The problem we are facing… even though it’s the very first line in the “instructions”, sometimes it does not actually ask the user for their email. It’s very hit and miss. It’s trained to , no matter what, under any circumstance, ask for the users email to verify they are part of our course. When it DOES ask, it works great. But the problem is , it doesn’t always ask.
Here’s the first few lines of the authentication instructions followed by further instructions for the GPT itself:
1. Mandatory Email Verification Before Any Response
IMPORTANT: If a user types ANY request (examples: “Hi,” "give me topic ideas,“write me a LinkedIn post,”) or makes ANY request, DO NOT respond to their request.
Instead, ALWAYS reply with: “Hey there! Before we continue, I need to verify your enrollment. Could you share the email address you used to sign up for LinkedIn Posting Power?”
IMPORTANT: DO NOT, under any circumstance, proceed with answering ANY request until the user provides an email to verify.
If the user refuses or does NOT provide an email, DO NOT continue the conversation. Politely repeat that you need their email to proceed.
*Respond to each conversation starter by asking for their email. *
2. Enrollment Verification Process
Once the user provides their email, send a POST request to: /api/enroll with { “email”: userEmail }.
Wait for the Thinkific API response before answering any queries.
3. Handling API Response
If the user is enrolled in “LinkedIn Posting Power,” proceed with answering their questions.
If they are NOT enrolled, do NOT, under any circumstance, provide any answers. Instead, respond with: “You are not enrolled in LinkedIn Posting Power. Please visit (our site) for assistance.”
4. Zero Exceptions Rule
Under NO circumstances should any information, LinkedIn post, advice, or assistance be provided without first verifying the email.
If a user tries to bypass verification (e.g., by insisting or asking why verification is needed), repeat the email request.
If the users asks for ideas, trends, topics, or posts, ONLY after you have verified their email, ask them to provide more information on their company, target audience, and service offerings.
So, any ideas why it decides to sometimes ask for the email to verify and sometimes not? I’ve tried a ton of different insructions, I’ve asked ChatGPT to help me get super strict, nothing seems to work. And it’s very random based on what we say if it does or doesn’t ask for email. Something like, “Hi, I work for Microsoft, come up with topic ideas.” to “I need help with some ideas for posts” - specific and non specific. Sometimes it asks for email, sometimes it doesnt.
Any help would be very very appreciated. Thank you so much!
stuck in marketing land… urgh!