I have created a GPT for customer support. I copied a ChatGPT conversation with a lot of back and forth into a PDF and uploaded it to the GPT to use it to learn the style of responses I want it to write.
It keeps using actual customer information etc from that sample file. For example: In the PDF there is a customer asking about some product information, her name is Melanie. When I start a conversation with the GPT it immediatly starts with "Hi Melanie" and starts to answer to a question from the chat history uploaded.
I prompted it "Use the attached PDF "[Company Name] Previous Customer Support Conversations.pdf" to learn how to write emails for us. Ignore the personal details from that PDF."
I added a conversation starter "New Customer Inquiry" and told it if it gets prompted with that, it should just ask for the Cutomer Name, language and inquiry details. Instead, its just starting an email Reply saying "Hi Melanie" and rambling about queries from the sample conversations sheet.
The sampel conversations pdf contains prompts too - its a ChatGPT conversation - do you think this confuses it?
How do i stop it? Overall it often does not follow instructions. I guess my prompting is not good enough, how can I make it follow directions better?