Custom GPT is not referencing knowledge source files on the first prompt

I have a custom GPT built for evaluation and have provided it with instructions to evaluate using a knowledge source file with a scoring rubric.

I have noticed today that the first prompt in a conversation does not use the knowledge source, but for any prompt after that, the custom GPT will reference the knowledge source file.

For example:

Prompt: evaluate this text and provide it a scoring tier: [text]
Response: Needs significant improvement
Prompt: evaluate this text and provide it a scoring tier: [same text]
ChatGPT shows that it is referencing a knowledge file.
Response: Above average

How can I get my custom GPT to reliably reference the knowledge source file in the first response, or do I need to begin the conversation before it has that capability?

3 Likes

Hi @owfox

I created a sample and simple Text Evaluation GPT to test for you.

I added one txt file about Scoring Rubric.txt
How it works properly as you need, you can see its instructions and content of file.

If you input simply ‘HAKUNA MATATA’ it will repeat instructions and file.

You can copy prompts from it what prompts you need, and add into your GPT.