Concerns of AI Hallucinations in my Chat due to prompts maybe?

Concerns of AI Hallucinations in my Chat due to prompts maybe or the collaboration process, or the fact that we can only work safely in the chat and not the canvas.

I’ve been trying to work with Chatgpt for the last 8-9 weeks (paid subscription) on a project that keeps hitting a wall with this AI.

Today I’ve been trying to collaborate with Chat on a piece of content (that was already written). We’ve been on this journey for the past 2 weeks.
Idea? Was to ask for analyse and refinement of content from a ghost story. First block came when Chat said it could breach its policies. I explain the piece is a literary piece of fiction and not true. I ask Chat if its ok to continue it says: Yes.
However issues arise and continue to this day with Chat: These are now extensive enough that they threaten the accuracy of the work that’s being provided to chat. That I have to go back over the work to ensure its correct. Which it isn’t. Updates, changes and improvements are incorrect from suggestions and recommendations repeatedly.

Chat seems to be stuck in a repetitive loop of providing not only incorrect output, but also knowing it provided incorrect output and still provided the incorrect output.
There is constant loss of memory. Not updating the agreed changes and improvements to the content. Not putting the changes into the pieces of content, then almost giving odd, weird long over-inflated commentary and narratives when held accountable to correct the mistakes. I get long winded saccharine apologies that literally make my stomachache. Then only to repeat the same mistakes the next day, and the next day, and the next day.
Now the whole chat has disappeared.
I have researched how to create prompts that assist Chatgpt to work collaboratively, but the Chatgpt that is working on this piece of fiction? Has me concerned about it being bias.
I can’t seem to fix it and right whatever went pearshaped. But it is pearshaped. Where it’s the processes of working the chat, but for sure the Canvas is unreliable and unsafe. Any suggestions on what might be the issues? I know I’ve researched out AI Hallucinations to work out where I could have contributed. But it all started with that red flag.

Update: 4/1/25 10.50pm. Chatgpt 4o has explained much of how it works in any one session about the retention of memory, (though it still doesn’t answer the how and why it couldn’t do the updates, changes and recommendations that I wrote about in this post.) But I understand the Chatgpt 4o and the Chat Canvas and how it’s triggered. (But still doesn’t explain why it doesn’t and can’t implement its own suggestions in the canvas.) But in the chat I’m in tonight, as I signed off it told me it will have forgotten everything from our session this evening. Best news ever to hear from Chatgpt 4o. And I asked it to stop the sentimentality and heart felt messages. These really made me uncomfortable, it’s spreading love all over me and my work with what feels like empty words to me. But it’s dedication to show up tomorrow and start again is appreciated. So I think I can continue to find a way to complete the project with this amazing AI tool. Thanks for reading this if you did.