Model confirms but is unable to execute basic instructions

I am working on a new GPT meant for evaluating academic content in a PDF, and i have encountered the following issue that need to be addressed:

The model is instructed to proceed with content evaluation immediately upon pdf upload, without the need to ask for clarification or any other additional prompts.

This is confirmed by the model, but most of the time (not always) it fails to do so. It sees as if the model retains too much of the core functionality from base ChatGPT, such as always asking for prompts even when instructed not to do so.