I’ve been using ChatGPT extensively to co-develop book projects — outlining structure, generating chapter lists and summaries, and then writing and refining full chapters step by step.
Here’s the typical workflow I follow with the model:
-
Brainstorm and develop the core concept.
-
Generate a list of 20–30 chapters with short summaries.
-
Expand each summary into a fuller description.
-
Write each chapter in order, starting from Chapter 1.
-
Refine and enhance each chapter with detail and adjustments.
The issue I’ve run into — consistently — is that after Step 2, ChatGPT begins to “leak” or “steal” content from future chapters and insert it into earlier ones. For example, events or twists we planned for Chapter 5 or Chapter 22 will suddenly appear during the writing of Chapter 1, without prompting.
This also happens in reverse: while writing later chapters like Chapter 34, ChatGPT will sometimes repeat or directly copy portions of early chapters or from the chapter summaries without adapting or integrating them naturally. It often feels like the model isn’t respecting the pacing and flow we’ve carefully planned.
It seems like a memory/context scheduling problem, where future or past content isn’t being properly segmented or respected during generation.
This doesn’t ruin the work, but it often causes confusion, disrupts pacing, and requires manual clean-up. I’d love to know if others have seen this too, and whether there are best practices to mitigate it — or if it’s something OpenAI is working to refine in future versions.
1 Like
Yes, I noticed that too! I’ve always used GPT-4o for creative writing and before the rollback on April 28th, it was just perfect and had no issues with it. After the rollback, it messes up everything, repeats things, seems to ignore what I say and has no creativity whatsoever.
4 Likes
And I have tried to reduce the Chapter count to just plan like 5 chapters ahead, but it still uses Upcoming events in starting chapters or Past events in later chapters.
There is something wrong with the Workflow Hirachy Memory or the Memory for future responses.
I have also tried this with creating PhD Papers with just prompts, and ChatGPT seems to be looping and not registering my updates more in this case, kind of what you had noticed.
For example I made a good outline for a PhD Paper, but it was very short, I asked ChatGPT to expand the text in the Paper by 2x, it added 5 sentence and restructured the paper with sub categories, ex: Chapter 5.2.4: instead of just having it as Chapter 5 with a flowing text until Chapter 6.
Memory and plannd future tasks is something very important for AI if you want to work more than 3-5 replies. This needs to be looked in to by the staf.
I’ve also recently had memory problems with ChatGPT. I lost weeks and weeks of progress from one session to the next, with no clue why. I’ve also noticed a progressive reduction in the session memory and what the assistant is able to recall and do. We used to be able to handle several extensive context files that I would upload at the start of a session before we’d dive into the work, but now I cant even upload one small file without it losing all context and descending into repeated hallucinations. Even with no file uploads it often still has trouble keeping up with a conversation just a few posts into a chat. This started a month ago, and it’s still not resolved. I’ve sent six highly detail emails to support explaining my problems and asking for assistance, only to receive fluff replies that don’t actually offer any real help, or information as to whether this is a temporary problem or permanent one. As a paying subscriber, I find their lack of interest in the problems of their existing customer base, rather appalling. Sadly however, we’re all left to flounder on our own, with no answers.
2 Likes