My new GPT is making up answers instead of using the dataset

I’m really new to this, just exploring. I tried making a new GPT and added a CSV dataset to its Knowledge then told the builder that the uploaded dataset contains information about books and the GPT should only refer to books listed in the dataset. The builder seemed to understand and confirmed my GPT would only refer to these books, but when I prompt the GPT to tell me about books it makes up random books that don’t exist instead of listing ones in the dataset. Am I doing this completely wrong? What’s going on? When I upload the CSV to ChatGPT I am able to get accurate responses to prompts based on the dataset, but not with the GPT I am trying to create. Help!

I’m having the exact same problem. My custom GPT is completely ignoring the attached TXT file most of the time but sometimes it works perfectly often. It will give a completely incorrect answer but claim that has gotten its data from the knowledge.


My experience;

Sometimes, if we save CSV files using Excel or other similar softwares, ChatGPT cannot recognize rows and columns, but as I said “sometimes”.

I save my CSV files putting in a TXT editor (I use Sublime Text).
So far, I never had any problems as many users faced.
But, maybe there are another better explanations from other users’ experiences.

I built a sample GPT for you to test how it works only using knowledge file without using “common reference” or “general knowledge”. I added a CSV file with 99 unique books. You may see it HERE and HERE, also you can see its instruction end of the chat. You can modify it for your needs.