Title: Please warn users BEFORE a long ChatGPT conversation reaches its context limit
I am requesting one very simple feature:
Please warn users before a conversation reaches the point where important earlier context may no longer be reliably available.
I am not asking OpenAI to remove context limits.
I am not asking ChatGPT to remember every conversation forever.
I am simply asking for a warning before it is too late.
For example:
“This conversation is approaching its context limit. Important details from earlier in the conversation may become unavailable. Consider creating a handoff summary or moving to a new chat.”
That single warning would have prevented hours of wasted work in my case.
What happened to me
I use ChatGPT for long-form YouTube production inside a Project.
For one video, I was working through approximately 37 separate items.
For each item, ChatGPT and I were building up production decisions such as:
-
which historical images to use
-
which images were rejected
-
the order in which the images should appear
-
what each image should communicate on screen
-
how the narration should be written to match those specific images
This was not casual conversation.
These decisions were the actual production data for the video.
I spent a long time building this up item by item.
There was no warning that the conversation was approaching a limit.
Eventually I had to continue in another chat within the same Project.
At that point, ChatGPT could no longer reliably reconstruct all of the detailed decisions from the previous conversation.
Worse, when I asked it to reconstruct the previous work, it sometimes filled missing information with plausible guesses and presented those guesses as if they were decisions we had actually made.
For example, for one section about the launch of Radio Tokyo Television in 1955, ChatGPT claimed that we had selected an image of “people watching a television receiver.”
We had never selected such an image.
When I challenged it, ChatGPT acknowledged that it had inferred a plausible image rather than actually recovering the previous decision.
That is a serious problem for production work.
The frustrating part: this was completely preventable
If ChatGPT had warned me before the conversation became too long, I could simply have said:
“Before we continue, create a complete handoff document containing every image selection, rejected option, display order, and narration decision.”
Then I could have started a new chat with that document.
Nothing would have been lost.
No reconstruction would have been necessary.
No hallucinated past decisions would have appeared.
And I would not have had to spend additional time re-checking dozens of items that had already been completed.
The problem is not the context limit
I understand that context windows have technical limits.
That is not what I am complaining about.
The problem is this:
The system knows that limits exist, but the user is given no meaningful opportunity to protect their work before those limits affect the conversation.
For users doing long-term creative work, research, coding, writing, planning, or production, this can destroy hours or days of accumulated decision-making.
The conversation may still exist in the sidebar.
The Project may still exist.
But the user only discovers that the working context has become unreliable after the damage has already occurred.
What I would like OpenAI to add
At minimum:
A visible warning when a conversation is approaching the point where earlier context may begin to become unavailable or unreliable.
Ideally, the warning could also include a button such as:
“Create handoff for new chat”
which would preserve:
-
confirmed decisions
-
unresolved issues
-
important constraints
-
selected/rejected items
-
current progress
-
next steps
But even without an automatic handoff feature, the warning alone would solve most of my problem.
I can create my own handoff document.
I just need ChatGPT to tell me when it is time to do it.
One final point
I remember earlier versions of ChatGPT giving me some indication when a conversation had become too long, while still leaving enough room to prepare for moving to another chat.
In my recent experience, I received no useful advance warning.
That change has made long-term production work significantly less reliable for me.
Please consider adding an advance context-limit warning.
This is a small UX feature that could prevent a very large amount of lost work for people who use ChatGPT as a serious long-term working partner.